Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

How language production shapes language form and comprehension

How language production shapes language form and comprehension HYPOTHESIS AND THEORY ARTICLE published: 26 April 2013 doi: 10.3389/fpsyg.2013.00226 How language production shapes language form and comprehension Maryellen C. MacDonald* Department of Psychology, University of Wisconsin-Madison, Madison, WI, USA Edited by: Language production processes can provide insight into how language comprehension Charles Clifton, University of works and language typology—why languages tend to have certain characteristics more Massachusetts Amherst, USA often than others. Drawing on work in memory retrieval, motor planning, and serial Reviewed by: order in action planning, the Production-Distribution-Comprehension (PDC) account links Gary Dell, University of Illinois at work in the fields of language production, typology, and comprehension: (1) faced Urbana-Champaign, USA Fernanda Ferreira, University of with substantial computational burdens of planning and producing utterances, language South Carolina, USA producers implicitly follow three biases in utterance planning that promote word order Joan Bresnan, Stanford University, choices that reduce these burdens, thereby improving production fluency. (2) These USA choices, repeated over many utterances and individuals, shape the distributions of *Correspondence: utterance forms in language. The claim that language form stems in large degree from Maryellen C. MacDonald, Department of Psychology, producers’ attempts to mitigate utterance planning difficulty is contrasted with alternative University of Wisconsin-Madison, accounts in which form is driven by language use more broadly, language acquisition 1202 West Johnson St., Madison, processes, or producers’ attempts to create language forms that are easily understood WI 53706, USA. by comprehenders. (3) Language perceivers implicitly learn the statistical regularities e-mail: mcmacdonald@wisc.edu in their linguistic input, and they use this prior experience to guide comprehension of subsequent language. In particular, they learn to predict the sequential structure of linguistic signals, based on the statistics of previously-encountered input. Thus, key aspects of comprehension behavior are tied to lexico-syntactic statistics in the language, which in turn derive from utterance planning biases promoting production of comparatively easy utterance forms over more difficult ones. This approach contrasts with classic theories in which comprehension behaviors are attributed to innate design features of the language comprehension system and associated working memory. The PDC instead links basic features of comprehension to a different source: production processes that shape language form. Keywords: language acquisition, motor control, language production, serial order, language comprehension, syntax, language typology, working memory a central role for experience in development and in adult perfor- INTRODUCTION Humans are capable of a remarkable number of highly com- mance. plex behaviors—we plan ahead, remember the past, reason, The statistical properties of the input have a similarly cru- infer, and invent. The origins of intelligent behavior are at cial role in some accounts of language use, including the role the core of classic debates in cognitive science concerning the of linguistic experience in acquisition (Hart and Risley, 1995) contributions of innate capacities and experience in the devel- and inadult comprehensionprocesses(MacDonald et al., 1994). opment of thought, perception, and action. For example, the However, the nature of the argument is critically different in fact that perception of motion in cardinal directions (verti- vision and in language. Visual experience reflects the nature of cal, horizontal) is superior to that in oblique directions has the physical world: Vision scientists do not need to explain why been attributed to the greater number of cells in visual cortex gravity creates many experiences of downward motion, and no devoted to processing cardinal motion directions than oblique one expects face perception researchers to explain why faces have ones (Rokem and Silver, 2009), and this result in turn is particular shapes. In language, however, the input to the perceiver thought to arise from visual experience: There are more motion is itself the consequence of language behavior—it is the utterances events in the world in cardinal directions than in oblique ones produced by other language users, who have their own cognitive (Dakin et al., 2005). Similarly, experience-based accounts of systems presumably shaped by their own experiences. This situ- face perception hold that face recognition behavior diverges ation lends potential circularity to experience-based accounts of from object recognition because perceivers’ visual experience language (Frazier, 1995), requiring solutions for two unknowns at with faces differs in critical ways from their experience with once: as in vision, language researchers must develop an account objects (Tarr and Gauthier, 2000). While such accounts don’t of the effects of experience on perception, but unlike in vision, deny innate factors in perception, they are notable in ascribing language researchers must also consider why the experience—the www.frontiersin.org April 2013 | Volume 4 | Article 226 | 1 MacDonald Language production shapes form and comprehension language—has the character it does. This difficult task is com- development and organization of plans for output sequences as pounded by the fact that the psycholinguists who study language “both the most important and also the most neglected prob- use are typically not the same people as the linguists who study the lem in cerebral physiology” (p. 114). He argued that complex nature of language form, so that there is a gulf between linguistic sequential actions such as speaking must be guided by a plan that theories of the nature of language and psycholinguists’ accounts is developed before execution, a view that continues to pervade of the effects of experience with language patterns. research in motor behavior, including language production. The This article is a step toward bridging this divide, offering construction of motor plans is a cognitively demanding activ- insight into both the origin of language form and also the effect ity; developing the utterance plan can be more demanding than of experience with these forms. The Production-Distribution- speaking itself (Kemper et al., 2011). The significant computa- Comprehension (PDC) account, first sketched in MacDonald tional difficulty of constructing and maintaining an utterance (1999) andelaboratedin workdescribed here, holds that the plan is a key component of the PDC, and so we consider these memory and planning demands of language production strongly planning operations in some detail. affect the form of producers’ utterances. Constraints imposed by the production process have two important consequences. DEVELOPMENT AND CARE OF THE UTTERANCE PLAN First, they contribute to understanding regularities in linguistic Language planning shares features of both high-level non- form: why languages exhibit particular properties, with differ- linguistic action planning and more fine grained motor control. ent frequencies across languages. Second, they determine many In high-level action plans, some elements have only loosely con- aspects of language comprehension. The claim is not that all strained sequences. In making coffee, an example extensively aspects of language form and comprehension can be traced to the discussed inresearchonactionplanning and control (Cooper computational demands of language production, but rather that and Shallice, 2000; Botvinick and Plaut, 2004), the coffee, cream, production’s impact in these areas is so pervasive that understand- and sugar can go into the cup in any order. Similarly, in some ing production becomes essential to explaining why language is (though by no means all) aspects of language planning, some the way it is, and why language comprehension works the way elements may be ordered in several ways, as in Jane bought a ham- it does. mer and some batteries at the hardware store, vs. At the hardware In this article I describe the Production, Distribution, and store, Jane bought some batteries and a hammer. Other aspects of Comprehension components of the PDC in that order, focus- action/motor plans are far more constrained—one must move ing particularly on lexico-syntactic phenomena. The section the hand to the coffee cup before grasping it, and in the case of entitled The First Step in the PDC: Production Difficulty and its language planning, there are language-specific constraints limit- Amelioration reviews the memory and control demands of lan- ing the range of permissible word orders, for example excluding guage production, producers’ attempts to mitigate them, and the hardware at store the. Thus language producers have word order patterns of word order, sentence form, and lexical-sentence pair- options in some cases but not others, and when there are options, ings that result. Findings in motor control, memory retrieval, producers must very rapidly settle on one form and inhibit oth- and short term maintenance suggest that many properties of ers from interfering, so as not to make speech error blends of language production that affect utterance form also arise in alternative forms such as some hammer and a batteries.This action and motor planning more generally. Next, the section behavior is an example of a winner-take-all process, and winner- entitled Distributional Regularities and Language Typology con- take-all neural mechanisms form an important part of accounts siders the effects of language production on language form and and computational models of both language production (Hartley views the potential contributions of the PDC in the context of and Houghton, 1996; Dell et al., 1997) and non-linguistic motor other accounts of why languages have some properties more behavior, including visual search (Ferrera, 2000) and the “syn- than others. Finally, ComprehensionConsequences inthe PDC tax” of birdsong (Jin, 2009). This winner-take-all property of addresses comprehension, showing that the PDC provides a dif- language production is critical in accounts of how producers acti- ferent framework for thinking about sentence comprehension vate the correct serial order of elements in articulation (Hartley and offers a different explanation of some classic results. and Houghton, 1996), and it provides our first example of how properties of motor planning affect distributional patterns in the THE FIRST STEP IN THE PDC: PRODUCTION DIFFICULTY language, in that this property affects the incidence of speech AND ITS AMELIORATION errors. Language production is a highly complex motor behavior, requir- The developing utterance plan must be maintained in an exe- ing the translation of conceptual information into an intricate cutable state as it is being developed. The plan is effectively “the sequence of motor commands to allow speaking, signing, writing, memory for what is to come” (Rosenbaum et al., 2007, p. 528), or typing. Although “production difficulty” and “motor con- with all the maintenance burdens of other short-term memories. trol” might suggest a discussion of articulation, here we consider Indeed, verbal working memory studies offer important insights difficulty arising in the development of the plan for the utter- into some of the memory demands of language production. In ance, well ahead of articulation . Lashley (1951) considered the both serial recall tasks (in which unrelated words are recalled in the same order they were presented) and language produc- 1 tion tasks (such as describing pictures), elements in the utterance “Utterance” here refers to all modalities (speaking, writing, signing). Each plan tend to interfere with one another, affecting the fluency of modality’s unique production demands should influence the distribution of forms in that modality, but those effects aren’t discussed here. speech. For example, phonological overlap among elements in Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 2 MacDonald Language production shapes form and comprehension the utterance plan increases the difficulty in both production and which include developing the plan, maintaining it, monitoring memory tasks (Acheson and MacDonald, 2009), and semantic the state of execution, and shifting attentional focus as the plan overlap between words increases errors in language production is executed over time. Second, they illustrate how speakers learn (Smith and Wheeldon, 2004) and memory tasks (Tse et al., 2011). implicit strategies to mitigate production difficulty, in this case Conversely, production of the correct serial order of elements learning to allocate more attention to the upcoming plan as they is improved by increased linguistic frequency or coarticulatory become more fluent, and learning to favor early execution and experience, both for memory tasks (Woodward et al., 2008)and incremental planning, with delaying tactics and additional dam- language production (Dell et al., 1997). Thus, production plan- age control if the plan runs out. And third, production-related ning has inherent working memory demands, with consequent learning affects the distribution of utterance forms that people interference and other pitfalls well known to memory researchers. produce, in this case the rate and distribution of speech errors and Because planning precedes execution, a key question in lan- pauses in utterances. The intersection of these last two points— guage production concerns the degree of advance planning before that the computational demands of language production can be execution begins. Language production is said to be incremen- mitigated, but with consequences for utterance form—will reap- tal, meaning that partial planning, execution, and subsequent pear below as a force in the distribution of syntactic forms in planning are interleaved. The scope of advance planning varies languages. in different circumstances and is at least partially under the MINIMIZING DIFFICULTY DURING PRODUCTION producer’s strategic control (Ferreira and Swets, 2002). Again, Incremental production—the interleaving of plan and production behavior is shaped by learned implicit strategies that execution—works only if new plan segments can be developed maximize fluency, as the scope of planning strikes a balance between competing demands. On the one hand, initiating exe- at a rate that keeps up with execution. New plan development in turn relies on retrieval from long term memory, and when cution before much planning is complete allows producers to begin speaking earlier, avoiding long pauses and retaining the this retrieval fails or requires extra time, production is delayed or derailed. We next review three memory-related production floor in a conversation. Early execution also avoids the mem- ory burden of maintaining and executing a large plan, as more biases that have substantial consequences for lexico-syntactic distributions in utterance form. complex plans require more time to initiate execution, both in speech (Ferreira, 1991) and in non-linguistic motor behaviors Easy First: a source of word order flexibility (Rosenbaum et al., 2007). However, interleaving planning and execution has the occasional negative consequence of the pro- As anyone who has been in a tip-of-the-tongue state knows, ducer finishing the executable portion of the plan before the next some words are more easily retrieved from memory than oth- portion is ready. Rather than letting everything grind to a halt, ers. This fact has enormous influence on language form, because speakers in this situation attempt to gain extra planning time by easily retrieved words and phrases tend to appear both earlier lengthening words or adding optional words and pauses, yielding in utterances and at more prominent syntactic positions (e.g., utterances such as “Have you seen theee ...um ...?” (Fox Tree sentence subject) than ones that are more difficult to retrieve and Clark, 1997; Ferreira and Dell, 2000). (Bock, 1982; Tanaka et al., 2011). An Easy First bias in incre- Beyond juggling planning and executing, language produc- mental production allows execution of utterances to begin early, ers must also keep track of where they are in the plan as it is starting with easily planned elements, leaving more time for being executed. Tracking the state of progress through the plan planning of more difficult ones. “Easier” (also termed more acces- is critical for avoiding repetitions, omissions and other sequenc- sible or available) words and phrases have been described as ing errors, but it comes at a cost, in that tracking plan progress more frequent, shorter (both number of words in a phrase, and itself carries substantial additional attention or maintenance bur- number of syllables in a word), less syntactically complex, more dens (Botvinick and Plaut, 2004). At the same time, the memory important or conceptually salient to the speaker, and previously for what has been uttered cannot remain too strong, because mentioned (“given”) in the discourse (Levelt, 1982; Bock and recently-executed actions can interfere with upcoming ones, lead- Warren, 1985; Tanaka et al., 2011). There are enough different ing to perseverations and other errors (Tydgat et al., 2012). The forces affecting ease of planning that the claims can seem cir- speaker must therefore balance the various subtasks in utterance cular: Easy entities are easy because they appear earlier in the planning in order to “activate the present, deactivate the past, and utterance. However, the essential claim—that utterance planning prepare to activate the future” (Dell et al., 1997, p. 123; a non- difficulty affects speakers’ choices of word order and sentence linguistic example is Deco and Rolls, 2005). An efficient allocation structure—gains external validity in several ways. First, difficulty of attention to past, present, and future is learned over time: stems from ease of retrieval from long term memory, and many Fluent adult speech reflects a bias toward the future, with compar- of the factors that promote early positioning in an utterance plan atively more anticipation errors (elements of the upcoming plan also predict the early positioning and accuracy of word recall incorrectly influencing the current execution) than perseverations in verbal memory tasks, including word length, frequency, con- of previously-uttered elements (Dell et al., 1997). By contrast, creteness/imageability, givenness, and other factors (Bock, 1982). young children, who are less experienced speakers, produce more Second, other action and motor planning processes show these errors overall and a relatively higher proportion of perseverations same Easy First tendencies. MacNeilage and Davis (2000) argued (Stemberger, 1989). The impact of these phenomena is three- that the distributional regularities of consonant and vowels in fold. First, they illuminate the demands of language planning, infants’ babbling and early words reflect infants’ tendencies to www.frontiersin.org April 2013 | Volume 4 | Article 226 | 3 MacDonald Language production shapes form and comprehension order segments more easily articulated within the infant vocal over time and over other intervening utterances. The effect is tract before more difficult ones. Similarly, research in navigation argued to be not (or not only) the temporary activation of recent shows an Easy First action ordering preference: when humans or plans but rather a manifestation of long-term implicit learning animals have to visit several locations along a path, they typically of syntactic structure (cf. Branigan et al., 1999; Chang et al., begin with the nearest one (Gibson et al., 2007); if that nearest 2006). On this view, language users are continually learning from one is made difficult to reach because of obstacles, then more their and others’ language use; with every utterance, a syntac- distant unobstructed locations tend to be visited first (Miyata tic plan becomes more likely to be used in the future. Thus, and Fujita, 2011). Similarly, when humans are describing routes while the phenomenon is often described as one of short-term through a network, they also tend to begin by describing the sim- repetition, its learning basis links it to retrieval from long term plest one first (Levelt, 1982). Third, Easy First biases in serial memory—whereas Easy First refers to the effect of retrieval of ordering inherently follow from computational models of action individual words on word order, Plan Reuse effectively refers to planning, in which alternative sub-plans compete for entrance the retrieval of the sentence structure itself. The two constraints into an action plan, via selection mechanisms in sequence plan- jointly exert their influence on utterance form: Even in lan- ning (e.g., competitive queuing, Grossberg, 1978; Hartley and guages with very free word order, allegiance to favored structures Houghton, 1996) or via gating functions of attention in models of (Plan Reuse) combines with Easy First in shaping utterance forms cognitive control, in which more practiced/easier elements, which (Christianson and Ferreira, 2005). require less attention, precede more difficult ones in a developing The reuse of at least partially lexically-independent abstract plan (Botvinick and Cohen, submitted). Thus, the Easy First bias plans is in some ways consistent with an autonomous syntac- in language production is not a stipulative principle or language- tic representation independent of semantics (Chomsky, 1957), specific phenomenon; instead it follows naturally from attested although the notion of adapting a prior syntactic plan to a new aspects of motor and action planning—that a plan precedes its utterance, and the notion of sentences, phrases, and words as execution, that planning is incremental, that the plan is hierar- plans and sub-plans, are less consistent with the contrast in gen- chical with subplans that must be ordered in some way, that plan erative linguistics between a stored lexicon vs. generative syntax. development entails retrieval from long term memory, and that Moreover, the reuse of abstract plans is not unique to language, this retrieval varies in speed and accuracy. as Plan Reuse appears in many non-syntactic and non-linguistic These results suggest that the way that utterance planning domains. Its effects are evident in recall from long term memory, unfolds over time has a substantial impact on the word orders in which people have a tendency to recall elements in the serial and sentence structures that language users produce. Moreover, order in which they have frequently occurred in past experience this work suggests a mechanistic basis for the observation that (Miller and Selfridge, 1950). There is also increasing evidence variation in language has functional importance (Givón, 1985): for structured non-linguistic stimuli such as action sequences Word order variation, such as active/passive forms (The noise affecting subsequent production of certain sentence structures, startled the boy vs. The boy was startled by the noise)and the suggesting that the re-use phenomena are not inherently linguis- English dative alternation (give Mary a book vs. give a book to tic (Allen et al., 2010; Kaiser, 2012). More broadly, similar Plan Mary) allows producers the freedom to place easily retrieved ele- Reuse appears in many non-linguistic motor behaviors in humans ments early, permitting early execution of the plan, and allowing and animals and is attributed to implicit motor learning. It is more time to plan the more demanding elements. Thus, in con- for these reasons that the reuse and adaptation of prior motor trast to Jackendoff’s (2002) suggestion that syntactic flexibilities plans for subsequent action is thought to be a hallmark of motor are vestiges of ancient protolanguage, before syntactic constraints planning and learning (Rosenbaum et al., 2007), and motor learn- became more rigid, the PDC holds that word order flexibility ing in the service of language appears to be no different. This has real value to language producers and emerges from action point reappears in the section Implications, Limitations, Future planning mechanisms that maximize fluency. Directions. Plan Reuse: a source of word order rigidity Reduce Interference Despite the enormous impact of Easy First on word order, it Whereas Easy First and Plan Reuse stem from ease of recall from cannot be the whole story—people’s utterances are not sim- long term memory, Reduce Interference reflects properties of ply strings of words ordered by ease of retrieval from memory. immediate memory instead of or in addition to long term recall. A Production also accommodates constraints on permissible word classic finding in verbal and non-verbal short-term memory tasks orders in a language. A second significant influence on utter- is that the to-be-remembered elements interfere with one another ance form also favors easy, more practiced plans, but in this in memory during the short interval between their presenta- case, what is easy is the abstract sentence plan itself rather than tion and recall, with increasing interference when the elements the word or phrase elements (sub-plans) within it. Producers share similarity in sound, meaning, spatial location, or other have a conspicuous tendency to reuse recently executed utterance dimensions (Conrad and Hull, 1964; Anderson, 1983). Because plans, so that the likelihood that a speaker utters a passive sen- utterance plans are maintained before execution, it is not surpris- tence, for example, increases if that speaker has recently heard, ing that elements in the plan also interfere with one another. For read, or uttered another passive sentence (Weiner and Labov, example, when two semantically related nouns must be planned 1983; Ferreira and Bock, 2006). This tendency toward Plan Reuse and uttered in close proximity (e.g., . . . the couch and the chair. . . ), (also called structural persistence or syntactic priming) persists utterances take longer to plan and contain more errors than when Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 4 MacDonald Language production shapes form and comprehension this similarity is not present (Smith and Wheeldon, 2004). These result from attempts to mitigate the computational demands of effects may be a consequence of winner-take-all production: The production planning rather than a specific discourse strategy to path from conceptual message to word selection includes the par- emphasize certain information for the comprehender, then we tial (unconscious) activation of many alternatives (couch, sofa, should also see effects of the third factor described above, Reduce loveseat, chair, furniture, etc.), and successful production requires Interference, interacting with the other two. This prediction is that only one of these enter the utterance plan. Having then set- also supported; Gennari et al. (2012) found that when the agent tled on couch and inhibited all others, the producer has additional and patient of an event are semantically similar (e.g. boy, girl), difficulty—interference—when it becomes necessary to retrieve people more frequently describe the patient in passive structures one of these inhibited options (e.g., chair; Tydgat et al., 2012). such as The girl was pushed (by the boy),in which theagent As with other examples discussed above, producers mitigate this (boy) is demoted to a by-phrase or eliminated entirely in agent- interference via choices of utterance form (Gennari et al., 2012). A less passives (The girl was pushed). Here the system mitigates the specific example is given in the next section, which considers how demands of production by omitting, delaying, or demoting sen- Reduce Interference interacts with Easy First and Plan Reuse. tence elements that are affected by memory interference. These results suggest that while producers may sometimes (consciously The three factors in action or unconsciously) select a syntactic structure such as passives Easy First and Plan Reuse can pull in opposite directions, as to convey a particular message, substantial variation in utter- Easy First promotes word order flexibility to allow easily-retrieved ance form stems from the degree to which certain choices can words before more difficult ones, whereas Plan Reuse promotes reduce production difficulty for the producer. The results also rigidity of word order via re-using previously uttered struc- suggest that both word order variation and word order rigidity tures. Cross-linguistically, many distributional patterns of word have real value in production planning. Both the tendency to lead orders reflect this tension, owing to different degrees of word with easy elements and the tendency to adopt well-worn sentence order flexibility in different languages. In English and many types emerge from the nature of learning and retrieval from long other languages, passive structures such as (1b) are more com- term memory, in that highly frequent elements or well-practiced mon with animate subjects (boy) than with inanimate subjects abstract plans, are preferred over more attention-demanding like window. This pattern follows straightforwardly from the alternatives. On this view, implicit choices of both lexical items greater ease of memory retrieval of animate nouns than inan- and sentence forms are shaped by the same memory-retrieval imate ones (Bock, 1982), and from the fact that the passive constraints. construction allows the easily-retrieved noun boy to be placed Beyond increasing the fluency of an individual’s utterances, early in the sentence and in a prominent position (sentence these three production biases have another important conse- subject). quence at the heart of the PDC that “individual-level behaviours result in population-level linguistic phenomena” (Scott-Phillips 1a. Active: The ball hit the boy hard, but he was OK. and Kirby, 2010, p. 1364). Summed over millions of utterances 1b. Passive: The boy was hit hard by the ball, but he was OK. and many language producers, implicit production choices favor- ing less-difficult forms create dramatic statistical regularities in The three production planning factors make testable predictions language usage linking conceptual messages, words, and sen- about variation in passive use. First, if the relationship between tence types. The next section relates this perspective to other noun animacy and active/passive form is the result of utterance approaches to language typology and universals and argues that a planning pulled between Plan Reuse (favoring the more com- greater attention to production processes offers insight into ques- mon Active form) and Easy First (favoring early mention of tions about why some distributional patterns are more frequent animates), then we would expect that animacy/Easy First effects than others. on structure would be smaller in those languages (such as Slavic languages) that have a strong bias to use active forms. Results DISTRIBUTIONAL REGULARITIES AND LANGUAGE of this type are perhaps not surprising, because by definition, TYPOLOGY a strong allegiance to a single dominant word order to convey Functional linguists, language typologists, and historical linguists a particular message will allow less room for word order flexi- investigate the distributional regularities across the world’s lan- bility to accommodate ease of retrieval (Myachykov et al., 2011; guages and their change over time, with one goal being the Gennari et al., 2012) . Second, utterance planning time should identification of significant cross-linguistic tendencies or lin- increase when these forces conflict compared to situations when guistic universals that could illuminate the nature of human they converge on the same form. This prediction is also supported 3 language . Many functional linguists point to language use as (Ferreira, 1994). Third, if these structure and word order choices a source of cross-linguistic patterns, meaning that languages tend to have (or develop over time) properties that serve the See Bresnan and Ford (2010), Stallings et al. (1998),and Wasow (1997),for needs of language users (see Bybee, 2006,for review). The other examples of the tension between Easy First and Plan Reuse, though not using these terms. The reasons why one language would have freer word order than another is of course something to be explained within any perspective. In This statement dramatically simplifies functional, historical, and typological production based approaches, large-scale corpus studies should prove useful, linguistics as well as the debate about whether there are truly universals of as in investigations of the rigidity of use of dative constructions in American language or merely asymmetries in the distribution of language features in vs. Australian English (Bresnan et al., 2007; Bresnan and Ford, 2010). the world’s languages (Evans and Levinson, 2009). www.frontiersin.org April 2013 | Volume 4 | Article 226 | 5 MacDonald Language production shapes form and comprehension PDC’s view is related, but it holds that specifically the needs of language production, memory recall, action planning, and motor producers have the most direct effect on patterns of sentence control. structure. Indeed, this article is by no means the first to sug- gest that utterance planning processes in language production PRODUCTION GETS TYPOLOGY WRONG have an important role shaping language typology and histori- In English and many other languages, shorter words and phrases cal change (e.g., Bock, 1982; Bock and Warren, 1985; Jäger and tend to precede longer ones, which has been attributed to Easy Rosenbach, 2008). Typologists have long observed that linguis- First (Stallings et al., 1998). However, Hawkins (2004) describes tic variation happens for a reason (e.g., Givón, 1985), and the some notable exceptions in Japanese and other “head-final” lan- production processes described above take steps toward a more guages, casting doubt on production as a source of typological mechanistic account of why word orders may vary from one sit- patterns of word order: “The preference for long before short uation to the next as a function the nature of retrieval from in Japanese is not predicted by current models of language pro- long term memory, the role of attention in gating competing duction, all of which are heavily influenced by English-type processes in utterance planning, memory interference among [languages]” (p. 110). Hawkins is correct that language produc- entities in the utterance plan, and other factors. Moreover, the tion researchers have investigated relatively few languages, and learning mechanisms of production that promote reuse of prior this concern is compounded by psycholinguists’ tendency to pur- plans are an obvious candidate for informing accounts of lan- sue narrow, controlled studies focusing individually on Easy First, guage change (e.g., Bybee and McClelland, 2005). Despite this Plan Reuse, or Reduce Interference, with relatively little atten- potential synergy between a more detailed study of language tion to the fact that multiple factors can contribute to retrieval production mechanisms and language typology, there is rel- from memory and thus word order. Things are improving on atively little consideration specifically of language production both these fronts, and more recent production accounts do inves- processes in the typology and universals literature (though see tigate the origin of opposing effects of phrase length in English Bybee, 2006;also Jäger and Rosenbach, 2008, and associated and Japanese (Yamashita and Chang, 2001), including a computa- commentaries). Why not? One obvious answer—lack of inter- tional model of language production that develops Short-before- action between language typologists and language production Long preferences when trained with English input and learns a researchers—is generally true but not fully satisfying, because it Long-before-Short preferences when exposed to Japanese input doesn’t address why researchers in these areas feel little motivation (Chang, 2009). Analyses of the model’s performance in the two to interact. Five more substantive assumptions underlying the dis- language environments point to Plan Reuse as one important connection are considered here, together with arguments for a force in developing the ordering preferences, in that tendencies rapprochement. for ordering object and recipient noun phrases reflect the adapta- tion of plans from more common sentences with only one noun phrase. Thus, production work may have been late to the party CONCEPTUAL REPRESENTATIONS EXPLAIN LANGUAGE FORM here, but if Chang is correct about the role of learning mecha- WITHOUT LANGUAGE PRODUCTION MECHANISMS nisms and Plan Reuse, then his mechanistic account of utterance Linguists have long noted that more conceptually salient ele- planning will prove central to these cross-linguistic differences. ments (i.e., those more important to the producer and/or comprehender) occur early in utterances, such as the ten- dency for agents/animate entities to appear before undergoers of THE GRAMMAR-PERFORMANCE DISCONNECT action/inanimate entities. These and related patterns have been The value of distinguishing linguistic competence (knowledge: attributed to varied forces, such as an Agent First principle in the grammar) and performance (use) has long been a source of Universal Grammar (Jackendoff, 2002) and functional accounts debate within linguistics (Newmeyer, 1998; Jackendoff, 2007)and in which elements that are salient in the discourse receive a promi- is beyond the scope of this article. Two trends are worth not- nent sentence position (Chafe, 1976; Goldberg, 2006). This latter ing, however: First, some linguistic approaches increasingly view position clearly shares a good deal with Easy First, but the pro- grammar itself as a graded representation emergent from experi- duction account goes beyond an appeal to salience in important ence with language tokens (Bybee, 2006; Bresnan et al., 2007), and ways. First, the Easy First bias in production grounds the con- this position (whether or not “grammar” is invoked in the expla- ceptual salience effect in ease of recall from long term memory. nation), is central to recent production-based accounts of word Since salience itself is not acting directly on prominence but order variation (Kuperman and Bresnan, 2012). Second, whatever rather via ease of recall, this approach correctly predicts that other one’s position on the nature of grammar, utterance planning itself non-salience factors affecting ease of recall (e.g., word length) clearly shapes utterance form, and as such, production merits can also affect word order. Second, the incremental nature of more attention in typology if only to better attribute distributions motor planning for production explains why the privileged loca- to the work of grammar vs. performance. tion for easily recalled entities is early in the utterance plan Relatedly, the culture of controlled laboratory studies in lan- rather than saving the easiest for last. And third, filtering con- guage production is at odds with typologists’ interests in the ceptual salience through the production system accounts for broad sweep of cross-linguistic patterns. These trends are chang- situations in which communicative goals influence salience (via ing in several ways. First, researchers are increasingly investigating task-specific allocation of attention) rather than the other way the link between individual-level phenomena (as studied in many around (Kuchinsky et al., 2011). Thus, “salience” does affect language production studies) and the population-level phenom- word order, but it gains external validity via an understanding of ena, where interactions among many individuals affect language Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 6 MacDonald Language production shapes form and comprehension form over time (e.g., Scott-Phillips and Kirby, 2010). Clearly the to make input more regular, as when a child uses holded for the PDC will benefit from improved understanding of individual- past tense of hold rather than held. Considered from the point population interactions. Second, there is increasing use of large of production, the use of holded vs. held is an implicit choice corpora as a form of production data being brought to bear on of utterance form over available options, similar to saying cat typological issues (e.g., Piantadosi et al., 2012), and a move by vs. kitty or a passive vs. active sentence. However, a child who some psycholinguists to adopt information theoretic approaches utters holded may have never encountered it before; what could to language performance, in which accounts of language distri- cause a bias toward producing a form that’s generally unattested butions invoke notions of communicative efficiency (e.g., Jaeger in the input? Ease of production is a good bet, both because and Tily, 2011). This perspective obviously has clear overlap with it is such a powerful force in adult language production and functional linguistic accounts of typology, but from the PDC per- also because child utterances are full of omissions and other spective, it’s important to incorporate a detailed account of the simplifications that reduce utterance difficulty despite being unat- cognitive control and memory demands of production planning tested in the adult input. The production force that promotes into accounts of “efficiency” and why some distributional pat- overregularization is Plan Reuse, where the abstract plan here is terns are more common than others. It remains to be seen how the regular inflection, which becomes increasingly common as this tension between information theoretic and more mechanistic the child learns more verbs (Rumelhart and McClelland, 1986). accounts plays out. On this view, children’s novel forms such as omitting an initial unstressed syllable (e.g., saying nana for banana)and holded for TYPOLOGY AND UNIVERSALS REFLECT ACQUISITION, NOT held are not distinct phenomena but are instead both examples PRODUCTION of the strong influence of Plan Reuse, reflecting the dominance Whereas many functional linguists consider adult language use of first syllable stress in English nouns (Echols and Newport, as key to understanding language universals and change (Bybee, 1992) and the dominance of the regular inflectional paradigm in 2006), others point to the child learner as the engine of change, English. either via the application of Universal Grammar (e.g., Lightfoot, This example argues for a different approach to considering 1999) or via innate learning biases in the child (Hudson Kam the role of learning in language typology and change, namely and Newport, 2009; Culbertson et al., 2012). From the PDC per- shifting the question from “Why are some forms more easily spective, this distinction between acquisition and use is a false learned by the child?” to a child version of the same question one, because learning in production (and comprehension) never we’ve asked about adults: “Why are some forms more often stops; there is not first a phase of acquisition and later one of produced?” Viewing the overregularization effect as owing to pro- use (Seidenberg, 1997; Chang et al., 2006) . Learning specifically duction choices is broadly consistent with accounts in which the in the service of production may be particularly important in effects of experience with individual words and with the regu- understanding the nature of language typology and change, first lar paradigm (the plan) vary with the amount of prior exposure because utterance planning requires memory retrieval, which is (Rumelhart and McClelland, 1986). This approach also yields the itself a powerful determinant of further learning (Karpicke, 2012). correct prediction that children who overregularize may nonethe- Thus, each production event should have on average a stronger less comprehend the irregular form that they don’t produce effect on subsequent learning than each comprehension event, (Clahsen et al., 2007; Ramscar and Yarlett, 2007). These results owing to the greater retrieval demands in production. Second, suggest there may be real value in considering the child’s utter- production is winner-take-all, meaning that someone who has ance planning demands in phenomena that have previously been perceived variable input in the past must commit to only one attributed to more general learning biases or Universal Grammar. form for any given utterance, with potential consequences both This position is clearly not anti-learning but rather an argument for subsequent distributional patterns and for learning over one’s for considering what the learning is for. own productions . Thus, more attention to learning for produc- tion could inform current unknowns concerning hypothesized LANGUAGE IS TAILORED TO THE COMPREHENDER links between acquisition and typology: “why learners acquire In arguing for a central role for production in shaping language certain types of patterns more easily than others (and why lan- form, the PDC does not deny that other forces may also influ- guages therefore more commonly exhibit these patterns)” (Aslin ence form. However, if these other forces turn out to be extremely and Newport, 2012, p. 174). powerful, they could erode the PDC’s claim for the centrality A central assumption in the role of learning biases in lan- of production in shaping language form. One alternative force guage acquisition and typology is that child learners are biased shaping utterance form is audience design,the idea that language producers tailor their utterances to accommodate the needs of An approach favoring continuity between acquisition and use raises ques- the comprehender. Clearly producers do make adaptations to the tions about discontinuities, specifically sensitive period effects in language needs of the perceiver; the act of language production is itself an acquisition. Proactive interference approaches, in which prior learning affects accommodation, in that the producer is adapting to the fact that the rate of subsequent learning (Seidenberg and Zevin, 2006), suggest that the perceiver is not a mind reader and needs an overt linguis- child vs. adult language users may have different effects on language form, but tic signal. However, this adaptation is inherently limited, because not because of a distinction between acquisition and use. 5 the producer is also not a mind reader and therefore cannot fully Frequency boosting, in which a slightly dominant form strengthens over time know what needs the perceiver has. Thus, both audience design (Singleton and Newport, 2004), should be a natural consequence of winner- take-all production and learning from one’s own utterances (Plan Reuse). and production-driven utterance choices likely exist in parallel. www.frontiersin.org April 2013 | Volume 4 | Article 226 | 7 MacDonald Language production shapes form and comprehension Attributing aspects of utterance form to producer needs (as is not whether there is audience design (there is), but rather in the PDC) vs. audience design is a complex undertaking, for how distributional patterns emerge from the specific computa- several reasons. First, the most obvious potential evidence for tional demands of language production as executed by producers audience design—that comprehenders benefit when producers who have a communicative goal. Because audience design con- use certain forms and find other forms difficult—turns out not to tributes to the computational demands of utterance planning, be that useful. Perceivers routinely benefit from statistical patterns researchers who study the mechanisms of production planning in their input that have no audience design; we can predict the tra- should accommodate theaudiencedesignliterature morefully, jectory of a bouncing ball, but the ball does not aim to help our and vice versa. tracking. Moreover, producer variation that might seem designed SUMMARY for an audience can instead have purely production-centered explanations. Consider the phenomenon of phonological reduc- This highly selective discussion contains almost no typological data and omits many issues in the current literature. The aim is tion, in which speakers introduce a word into a discourse with a fairly careful articulation and later re-mention it in a less precise not to review PDC contributions to typology (that section would be very short indeed) but rather to suggest that there is sufficient “reduced” form (Fidelholtz, 1975). Listeners clearly have learned these reduction patterns and benefit from them (Dahan et al., promise for cross-disciplinary interaction, specifically that the computational demands of utterance planning, and producers’ 2002), but this benefit does not mean that phonological reduction is designed to help the listener. Indeed, reductions of this sort are attempts to minimize them, should be investigated further as an important driving force in cross-linguistic language typology and an inherent consequence of motor learning and practice gener- change. Though not elaborated here, the reverse is also true: Work ally, including in motor behaviors with no audience (Müller and Sternad, 2004). on the statistical distributions in the world’s languages, and the way that languages change over time, can inform psycholinguistic A second complicating factor is that audience design is not cost-free: accommodating a comprehender will itself impose accounts of language processes (e.g., Feist, 2010; Culbertson et al., 2012). demands on the producer. The producer must work to iden- tify perceiver needs, and as this task becomes more difficult (e.g., requiring more elaborate inferencing), the amount of audi- COMPREHENSION CONSEQUENCES IN THE PDC ence design in the utterance declines (Horton and Keysar, 1996; Having reviewed implicit choices of utterance forms and con- Bard et al., 2007). Thus, we can’t talk about whether utterance sequences for distributions in the language input, we now con- form comes from perceiver or producer needs, because perceiver sider what comprehenders do with these distributions. The next accommodation inherently creates demands on the producer. section addresses what language users learn about distributional Third is the existence of accommodation in the other direc- regularities, and the two sections after that review two classic tion, in that perceivers accommodate the needs of the speaker examples of sentence comprehension phenomena, for which pop- (Duran et al., 2011). Some obvious examples, many of which are ular theories have attributed comprehension behavior to archi- not always identified as producer accommodation, include vari- tectural properties of the comprehension system—in effect, that ous forms of phonetic adaptations (Kraljic et al., 2008)and ambi- comprehension works the way it does because innate parsing guity resolution at lexical, syntactic, and other levels (MacDonald biases make it so. By contrast, the PDC approach suggests that et al., 1994). Perceivers may often be good at accommodation the comprehension results reflect distributional regularities in the because they have direct information about the producer’s per- language, which themselves can be traced to the joint actions of spective from the utterance itself. Indeed, recent information- Easy First, Plan Reuse, and Reduce Interference shaping the forms theoretic analyses have suggested that overall communicative of utterances during utterance planning. efficiency is not optimal when the producer is maximally clear and redundant, which would make utterances longer and more DISTRIBUTIONAL REGULARITIES AND PREDICTION IN carefully articulated than the perceiver needs. Instead, commu- COMPREHENSION nicative efficiency is higher when the producer uses short ambigu- Linguistic signals unfold over time, creating long distance depen- ous words and permits phonological reduction and substantial dencies, where the interpretation of some input is dependent additional ambiguity (Piantadosi et al., 2012). This arrangement on previous or upcoming parts of the signal. Integrating over works because comprehenders are so good at ambiguity resolu- these dependencies involves use of probabilistic information in tion and other forms of speaker accommodation. Such results both forward and backward directions to settle on the most likely turn the notion of audience design on its head: Tuning the conver- interpretation of the input. In the backwards direction, recently- sational interaction primarily to the producer’s needs, and letting encountered information allows further refinement of the earlier the perceiver accommodate the producer, is in a broad sense a input (MacDonald, 1994), with some effects strong enough to form of audience design: The producer adopts utterance forms affect what perceivers report they hear (Warren and Sherman, mitigating difficulty and maximizing fluency so that the conversa- 1974; Connine and Clifton, 1987; Mack et al., 2012). Use of statis- tion proceeds efficiently, without bogging down the process with tical dependencies in the forward direction does not necessarily more redundancy than the perceiver needs. entail exactly predicting upcoming words but rather generating These considerations suggest that audience design is not general expectations about grammatical category, gender, and incompatible with producers’ implicit choice of utterance forms other properties that greatly narrow the scope of possibilities that mitigate production difficulty. On this view, the key question (Van Petten and Luka, 2012). The notion that comprehenders are Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 8 MacDonald Language production shapes form and comprehension generating expectations for upcoming input has been a contro- This approach accorded well with the fact that, with few excep- versial one, as ithas notalways been clear thatpredictions could tions (Fodor and Inoue, 1994; Altmann et al., 1998), the lexi- be sufficiently constraining or efficiently computed (MacDonald cal content of sentences like (2) has minimal effect on English and Seidenberg, 2006). However, partial predictions can emerge speakers’ strong bias in favor of local modification, making verb naturally from a system operating under time pressure (Allen modification ambiguities the best available evidence for lexically- and Seidenberg, 1999), and predictions may arise from many independent innate parsing algorithms. correlated language statistics, so that low levels of the linguis- As Table 1 summarizes, the PDC approach accounts for the tic signal, such as acoustic or orthographic form, can provide local interpretation biases without innate parsing algorithms. extremely early probabilistic information about grammatical cat- Instead the effects stem from learning over the distributional reg- egory or syntaxeveninadvance of word recognition (Dikker ularities in the language, which in turn stem from the biases of et al., 2010). Moreover, parts of the signal that have predictive producers to favor certain sentence forms that minimize produc- value for upcoming percepts not only speed the processing of tion difficulty. the predicted elements downstream but may themselves be pro- In Step 1 in the table, the Easy First production bias dis- cessed more rapidly than uninformative signal, owing to cortical courages production of distant modification sentences like (2c) feedback mechanisms gating attention toward potentially infor- because more easily planned alternatives exist. In (2c), a relatively mative input (O’Brien and Raymond, 2012). Together, this work long phrase (that his cousins left)precedes a shortone (yester- reflects a point that’s evident in information theoretic accounts day), but Easy First promotes a short-before-long phrase order, of language processing but hasn’t consistently penetrated other as in John said yesterday that his cousins left,or Yesterday, John comprehension approaches, that there is always ambiguity in the said that his cousins left. Step 2 identifies the distributional con- language signal as it unfolds over time, and uncertainties about sequences of speakers avoiding utterances like (2c): Ambiguous both the upcoming and the recently encountered signal are a sentences like (2a) typically have a local modification interpre- source of processing difficulty (e.g., Hale, 2006; Levy, 2008). Use tation like (2b). Comprehenders are extremely sensitive to these of distributional regularities to reduce this uncertainty is a pow- statistics (Step 3), and they have difficulty comprehending largely erful advantage in comprehension, and the next two examples unattested forms like (2c), but they readily comprehend a spe- suggest how a deeper appreciation of this fact, together with an cial type of distant modification sentences that don’t violate Easy understanding of how production processes create certain dis- First and that do exist in the language. These results suggest that tributional regularities, reframes our understanding of sentence rather than an innate comprehension bias for local modification, comprehension. perceivers have a learned bias toward what has happened in the past, and that this prior linguistic experience owes to aspects of production planning. REINTERPRETING PARSING PRINCIPLES: VERB MODIFICATION This claim for the role of past experience on subsequent com- AMBIGUITIES prehension processes is at the heart of constraint-based accounts A pernicious type of sentence ambiguity, the verb modification of language comprehension, which have been applied to many ambiguity, is shown in (2), in which an adverbial phrase could modify one of two different actions described in the sentence. Example (2a) shows a fully ambiguous structure, (2b) shows an Table 1 | Production-Distribution-Comprehension (PDC) account of example in which verb tense disambiguates the sentence in favor greater comprehension difficulty for ambiguities resolved with of a the local modification interpretation in which the adverb yes- distant modification (2c) than with local modification (2b). terday modifies the nearest verb left rather than the more distant phrase will say, and (2c) is an example of distant modification, in PDC STEPS which tomorrow modifies the distant verb, will say. 1. Production: Easy First, where shorter phrases precede longer ones, discourages production of ambiguous structures like (2a) with intended distant modification, and instead promotes production of 2a. Verb Modification Ambiguity: John said that his cousins left other forms to convey the same message (MacDonald, 1999; yesterday. MacDonald and Thornton, 2009). 2b. Local Modification: John will say that his cousins left yester- 2. Distribution: As a result, ambiguous sentences with intended distant day. modification are much rarer than ambiguous sentences resolved with 2c. Distant Modification: John will say that his cousins left local modification (Sturt et al., 2003; MacDonald and Thornton, 2009). tomorrow. 3. Comprehension: The comprehension patterns reflect the language statistics in Step 2: (a) Overall, the rarer distant modifications are harder than the more English comprehenders greatly favor local modification (2b) over common local modification sentences (Altmann et al., 1998; distant modification (2c). This pattern is often thought to arise MacDonald and Thornton, 2009). directly from innate parsing or memory biases to favor local (b) However, a subtype of verb modification ambiguities don’t violate phrasal relationships over long distance ones, variously formu- Easy First in their distant modification form, owing to the relative lated as Right Association (Kimball, 1973), Late Closure (Frazier, length of phrases in these sentences. These are readily produced 1987), and Recency (Gibson et al., 1996). A key assumption by speakers who intend distant modification, are common in the has been that these parsing principles operate on purely syntac- language, and are easily comprehended (MacDonald and Thornton, tic representations without lexical content (e.g., Frazier, 1987). 2009). www.frontiersin.org April 2013 | Volume 4 | Article 226 | 9 MacDonald Language production shapes form and comprehension other syntactic ambiguities (see MacDonald and Seidenberg, 3b. Subject relative: The woman [who yelled at me] said I’d have 2006, for review). The added value of the PDC is (a) a greater to pay for the broken glass. emphasis on the role of learning probabilistic constraints, and (b) an account of the production basis of the language distributions The origin of relative clauses’ importance can be traced to claims that people learn and use to guide comprehension. Extending by Miller and Chomsky (1963) concerning reasons behind the the PDC to other syntactic ambiguities is ongoing; the approach comparative difficulty of subject relatives vs. center embedded holds promise because (a) these ambiguities turn on the rel- object relatives. Chomsky and Miller (1963) observed that the ative frequency of alternative uses of language, which can be repeated recursive operation of embedding one object relative readily learned from input (Wells et al., 2009), and (b) certain inside another one yields an uninterpretable sentence; their exam- production choices are known to affect syntactic ambiguity. For ple was The rat [the cat [the dog chased] killed] ate the malt. Miller example, variation in availability of genitive forms (the profes- and Chomsky (1963) viewed the difficulty of these sentences sor’s daughter vs. the daughter of the professor)in English vs. as following from a distinction between linguistic competence other European languages affects the distribution of noun mod- and ability to use that knowledge—linguistic performance. They ification ambiguities and their interpretation in these languages argued that while linguistic competence (here, recursion) is infi- (see Mitchell and Brysbaert, 1998,for review and Thornton nite, performance, specifically the ability to use this knowledge to et al., 1999 for constraint-based studies of cross-linguistic simi- comprehend center embedded structures, is constrained by lim- larities and differences). Similarly, producers manage production itations on short-term memory capacity (Miller, 1956). In the demands through the use of optional words (e.g., Ferreira and case of object relative clauses, the memory burden stems from Dell, 2000), which have substantial effects on ambiguity, the dis- the multiple incomplete noun-verb dependencies arising as the tribution of form-meaning pairings, and consequent experience- sentence unfolds, so that the comprehender must first anticipate driven ambiguity resolution processes. Thus, the PDC prediction a verb for each noun (the rat the cat the dog)and hold these is that all syntactic ambiguities can ultimately be traced back unintegrated nouns in memory, and then when the verbs are to producers’ implicit utterance choices (many in the service of encountered (chased killed ate), associate them appropriately with reducing utterance planning difficulty), the consequent distribu- the nouns (Wanner and Maratsos, 1978; Gibson, 1998). By con- tions in the language, and comprehenders’ learning over those trast, the more comprehensible English subject relatives interleave distributions. nouns and verbs, reducing the memory burdens: The dog [that chased the cat [that killed the rat]] howled. REINTERPRETING SYNTACTIC PARSING AND WORKING MEMORY It is difficult to overstate both the impact of Miller and BURDENS Chomsky’s analysis and the subsequent reach of relative clauses The next example, relative clause interpretation, repeats the PDC into virtually all corners of language comprehension research. argument—mitigating production difficulty leads to utterance Several additional factors have contributed to the central posi- choices that lead to distributional regularities that lead to com- tion of relative clauses in theories of memory and language use. prehension patterns. Relative clauses nonetheless merit detailed First, relative clauses are widely held to be syntactically unam- attention, first because they illustrate complex interactions of all biguous (Babyonyshev and Gibson, 1999), so that comprehension three production biases, and second because they have played an difficulty can’t be attributed to ambiguity resolution processes. outsized role in theories of both syntax and language compre- Second, subject and object relatives can be made to differ by hension, so that a revision of traditional claims has substantial only the order of two phrases, as in the order of the senator consequences. and attacked in (4a,b), so that researchers can contrast com- prehension of sentences for which the lexical content seems perfectly matched. The vast majority of a very large number The relative clause trifecta: recursion, competence-performance, and working memory of studies in English and many other languages, across chil- dren, adults, individuals with brain injury, disease, or devel- Relative clauses are noun modifiers that include a verb, as in opmental atypicality, show that object relatives are more dif- examples (3a,b). In (3a) the ball is being modified by the brack- ficult than their matched subject relatives (see O’Grady, 2011, eted relative clause; because the ball is the object of the relative for review). The logic here seems perfectly clear: Because the clause verb (threw), this structure is called an object relative (or difference in difficulty can’t be ascribed to lexical factors or center-embedded) clause. A subject relative clause is illustrated in ambiguity resolution, it must reflect purely syntactic opera- (3b), where woman is the subject of the relative clause verb yelled. tions and the memory capacity required to complete them These two examples seem pretty innocuous, but in fact subject (Grodner and Gibson, 2005). and object relative clauses have played a central role in defining the differences between language competence and performance in generative linguistics, and they also have had an enormous 4a. Object relative: The reporter [that the senator attacked] impact in essentially every area of comprehension research, from admitted the error. acquisition, to adult comprehension, to studies of aphasia and 4b. Subject relative: The reporter [that attacked the senator] other language impairments. admitted the error. 3a. Object relative: The ball [that I threw to Harold] went over This competence-performance account of working memory his head and broke a window. overflow in relative clause comprehension continues as the Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 10 MacDonald Language production shapes form and comprehension Table 2 | PDC account of greater comprehension difficulty for object dominant perspective in linguistics, language acquisition, adult than subject relative clauses (citations refer to English results). psycholinguistics, and communicative disorders, despite crit- icisms of each of the components of this argument. These 1. Object relatives (5a) are common when the noun being described is criticisms include evidence that multiply center-embedded sen- inanimate (toy) but are avoided when the relative clause describes tences need not be incomprehensible (Hudson, 1996), com- something animate (boy), passive relatives (5b) are produced instead prehension difficulty is strongly influenced by the words in (Montag and MacDonald, 2009; Gennari et al., 2012). These patterns the sentence and therefore cannot reflect purely syntactic pro- owe to at least three production biases: cesses (Traxler et al., 2002; Reali and Christiansen, 2007), (a) Easy First: animate nouns are conceptually prominent and easily object relatives do contain a non-trivial amount of ambigu- retrieved from memory, leading to their position in early or ity directly related to comprehension difficulty, again refut- prominent sentence positions. The passive relative (5b) allows the described noun to be in the prominent subject position of the ing the assumption that relative clauses provide a pure mea- relative clause. sure of syntactic difficulty (Gennari and MacDonald, 2008), (b) Plan Reuse: the rate of passive relatives varies with the viability of the degree of prior experience with object relatives predicts passives in the language more generally, reflecting the reuse of comprehension success in children and adults, a result not passive forms from other sentence types (Montag and captured by memory overload approaches (Roth, 1984; Wells MacDonald, 2009). et al., 2009), people’s comprehension capacity for recursive struc- (c) Reduce Interference: there is more interference between tures is more accurately described by a system in which work- conceptually similar entities [e.g. boy/girl in (5)] than when an ing memory is inseparable from linguistic knowledge than by animate entity (girl) acts on an inanimate one (toy). This one with separate competence and performance (Christiansen interference can be reduced by omitting the agent in the utterance and Chater, 2001), and that cross-linguistically, relative clause plan, which is possible in passive forms (5b), but not in object complexity does not always predict comprehension difficulty relatives (5a). The higher the conceptual similarity between (Lin, 2008; Carreiras et al., 2010). Theresilienceof mem- sentence participants in the event to be described, the more speakers produce passive agent-omission relative clauses ory overflow accounts in the face of these myriad challenges (Gennari et al., 2012). in part reflects the essential usefulness of the constructs of 2. People readily learn these correlations between animacy and relative working memory capacity and competence-performance dis- clause type (Wells et al., 2009). tinctions in cognitive science. However, a second factor is that 3. Comprehenders who encounter the start of a relative clause have there has been no really compelling alternative account that very different expectations for how it will end, depending on whether captures both the subject-object relative asymmetry as well as something animate or inanimate is being described, with these other phenomena. The PDC approach aims to provide consequences for comprehension: exactly this. (a) When relative clauses describe something inanimate like toy, English speakers rapidly anticipate an object relative (5a); for The relative clauses that people produce animates (boy), object relatives are vanishingly rare and are not Insight into why object relatives are hard requires noting pro- expected by comprehenders (Gennari and MacDonald, 2008). ducers’ available choices, specifically that there are two ways to (b) The less producers are willing to say an object relative to convey a describe the patient/theme of some action with a relative clause, particular message, the less comprehenders expect one, and the either an object relative (5a) or a passive relative (5b; curly brack- more difficult the comprehension is when a sentence in fact turns out to contain an object relative clause (Gennari and MacDonald, ets indicate the optional passive “by-phrase” identifying the agent 2009). of the action). 5a. Object relative: The boy/toy [that the girl splashed] was are produced less often when describing animate entities than dripping wet. 6 inanimate ones . 5b. Passive relative: The boy/toy [that was splashed {by the girl}] Figure 1 also shows large cross-linguistic differences in the was dripping wet. overall rate of object relative use. The reasons behind these differ- ences are quite complex and of course reflect important topics in Step 1 in Table 2 describes how producers’ use of object rel- language typology. Some variation in overall tolerance for object atives vs. passive relatives is shaped by the joint action of relatives appear to reflect Plan Reuse and the viability of passives Easy First, Plan Reuse, and Reduce Interference. When English in main clauses in a language (Montag and MacDonald, 2009), producers are describing something inanimate (e.g., toy), they and other important factors may include whether the language readily produce object relatives like (5a), but they almost has other utterance forms that speakers might use beyond object never do this to describe something animate (boy). Instead, they utter passive relatives like (5b). This pattern is not lim- In all Figure 1 studies, native speakers answered questions about cartoon pic- ited to English; my colleagues and I have investigated rela- tures in which animate agents acted on animate or inanimate entities (e.g., a tive clause production in six languages, which differ widely girl in a pool splashing a boy or a toy). Critical questions required speakers in word order in main and relative clauses, the amount of to describe the objects of actions; e.g., What is green? referred to the toy being case marking on nouns, the availability of alternative struc- splashed. Participants were not explicitly instructed to use relative clauses but tures to express this same message, and many other proper- frequently did so, with replies such as The toy that the girl is splashing/that’s ties. Figure 1 shows that in all six languages, object relatives being splashed {by the girl}. www.frontiersin.org April 2013 | Volume 4 | Article 226 | 11 MacDonald Language production shapes form and comprehension On this view, relative clauses, which have been central to cur- rent conceptions of memory and language use in virtually every subfield of psycholinguistics, turn out to be wholly unsuited for that role, as they are not unambiguous, and their comprehen- sion reflects detailed knowledge of correlations between words and structures, not abstract syntactic representations. What then becomes of working memory limitations as a source of com- prehension difficulty, particularly within Miller and Chomsky’s (1963) competence-performance claims for infinite recursion limited by working memory? The short answer is that researchers may further debate competence-performance distinctions, but relative clauses should no longer be offered as evidence of over- flow of syntactic memory representations that limit infinite recur- sive capacity. A more precise answer about implications of the FIGURE 1 | The frequency with which object relative clauses are relative clause work requires closer attention to what working produced to describe animate and inanimate entities in a picture memory is and isn’t. In saying that the PDC account refutes description task, calculated as a percentage of all relative clauses produced. The English, Spanish, and Serbian data are from Experiments claims for working memory limitations in sentence comprehen- 1a,2,and3of Gennari et al. (2012), respectively. The Japanese data are sion, my colleagues and I do not mean that working memory from Montag and MacDonald (2009), Korean from Montag et al. doesn’t exist—to the contrary, a prime reason why language users (in preparation), and Mandarin from Hsiao and MacDonald (in preparation). track the statistics of the language and use them to anticipate upcoming input is precisely because language comprehension and passive relatives, and the extent to which nouns are marked requires significant memory capacity, and generating expecta- for case and the flexibility of word order in a language, both tions for likely outcomes reduces these burdens. However, we of which appear to modulate the degree interference between do reject the notion that people’s working memory capacity can the agent of the relative clause and the entity being described be described as a performance limitation independent of their by an object relative clause (Gennari et al., 2012). There are linguistic knowledge/competence (MacDonald and Christiansen, undoubtedly other complex influences as well. 2002; Acheson and MacDonald, 2009; Wells et al., 2009). Our Although we are just beginning to understand the factors position reflects broader trends linking working memory and behind the patterns in Figure 1, it is clear that speakers’ very dif- long-term knowledge (Cowan, 2005), emergent from the tem- ferent choices for animate-describing and inanimate-describing porary maintenance needs of other cognitive processes (Postle, relative clauses have robust effects on the distributional regular- 2006). Specifically for relative clauses, comprehension capacity ities in these languages. Steps 2–4 of Table 2 show the cascade varies with long term knowledge of these structures, derived from of consequences of these choices—comprehenders rapidly learn experience. Language producers provide some kinds of experi- the robust form-meaning correlations (Step 2), and they bring ences (some kinds of relative clauses) more than others, with this information to bear in comprehension, such that they consequences for language distributions, learning over those dis- expect object relatives where they’re commonly produced but tributions, and for the memory demands needed to comprehend are surprised by them in unexpected environments, leading to these structures—the memory capacity and experience cannot be comprehension difficulty (Step 3). The vast majority of studies separated. Of course computational limitations, including mem- demonstrating object relative processing difficulty have used ory limitations, are also at the heart of the PDC argument for materials in which something animate is being described—the why producers prefer some utterance forms over others, but this very situation that producers avoid and that comprehenders does not mean that the competence-performance distinction can have learned not to expect. Together, the steps in Table 2 suggest simply be shifted to production, because again, linguistic work- that object relative clause comprehension is simply another ing memory, specifically the capacity to produce certain utterance example of ambiguity resolution—comprehenders are “led down forms, is not separate from long term linguistic knowledge or the garden path,” as the saying goes in parsing research, by experience (Acheson and MacDonald, 2009). relying on past experience that leads to incorrect expectations for these unusual sentences, and the results do not reflect any pure SUMMARY effect of syntactic complexity on comprehension (Gennari and The two cases reviewed here, verb modification ambiguities and MacDonald, 2008) . relative clauses, exemplify the PDC’s point that an understanding of production choices in a language is critical for understand- ing comprehension. That idea has been implicit in non-syntactic A fuller treatment than is presented here would include the fact that object relatives with pronoun embedded subjects (The boy/toy she splashed....) comprehension work for decades (e.g., in lexical frequency effects have different production biases, different rates of production, and differ- on word recognition, in that frequency is inherently an effect ent comprehension patterns than the examples discussed here. We must of experience and ultimately producers’ word choices), but it’s also consider whether Easy First, Plan Reuse, and Reduce Interference pro- vide an adequate account of why multiply-embedded object relatives, like Miller and Chomsky’s (1963) The rat [the cat [the dog chased] killed] ate the difficulty here can also be traced to ambiguity resolution gone awry rather malt, are essentially never produced, and the extent to which comprehension than hard limits on working memory capacity. Frontiers in Psychology | Language Sciences April 2013 | Volume 4 | Article 226 | 12 MacDonald Language production shapes form and comprehension quite another thing to claim that we could abandon many of the This linkage between action planning and the mechanisms special-purpose syntactic interpretation mechanisms common in of language production has several intriguing implications the parsing literature if we understood sentence production bet- for the way language researchers view language form and ter. It will take some time to test this view in other constructions use. First, an implication for psycholinguistics: For decades, and languages, but in the meantime, the availability of exten- experience-based sentence comprehension research has empha- sive language corpora in many languages permits comprehension sized the non-independence of lexical and syntactic represen- researchers to examine the relationship between production pat- tations during the comprehension process (e.g., MacDonald terns (in the corpus) and comprehension behavior, even if they et al., 1994). By contrast, language production and motor/action have not yet investigated the production pressures that create planning more generally rely on abstract high-level plans the distributional regularities that are observed in a corpus. The that appear quite independent from the elements in the PDC suggests that it is essential to investigate such linkages before plan. Understanding how the demands of comprehension declaring that comprehension behavior owes to highly specific and production integrate lexical and more abstract hierarchi- design features in the language comprehension system. cal representations is an important challenge as these fields move forward. One possibility is that comprehension pro- IMPLICATIONS, LIMITATIONS, FUTURE DIRECTIONS cesses may draw on covert language production processes and The PDC begins with something utterly uncontroversial, that other aspects of non-linguistic motor planning (Pickering and language production is hard. The next step is no less obvious Garrod, 2007). If so, production may be doubly intertwined to production researchers, that language producers try to make with comprehension, both in the PDC’s view of produc- things easier, and that their attempts affect the form of the utter- tion mechanisms generating the statistics of language forms ances they produce. From there we get into somewhat more that drive comprehension, and also Pickering and Garrod’s controversial territory that (a) producers’ choices of utterance argument for covert production processes in the service of forms, repeated through the population, have a significant role in comprehension. explaining language typology and change over time, and (b) lan- Second, the link between language and action planning has guage users learn these statistical patterns and rapidly use them implications for how we view language itself. An enormous lit- to interpret new input. There are aspects of all of these ideas in erature considers how language is distinct from non-linguistic the literature, but the PDC is greater than the sum of these parts cognition (see Newmeyer, 1998; Jackendoff, 2002, 2007,among in suggesting that the downstream influences of production pro- numerous others), but the PDC may be able to contribute cesses are so strong and so pervasive that we must take production to the discussion. There has been little work to date investi- processes into account in developing theories of language form, gating the commonalities and differences between the abstract change, and comprehension. hierarchical plans that underlie sentence production and those One of the ways that the PDC is different from related ideas that underlie non-linguistic motor behavior. To the extent that is its emphasis on a specifically mechanistic account of language such commonalities exist, they could suggest that syntax, at production. It is certainly not wrong to appeal to more abstract least as it is realized in creating utterances, has a potential notions of communicative efficiency in accounting for producers’ homologue in non-linguistic systems and therefore is not some- choices of utterance forms (e.g., Jaeger and Tily, 2011; Piantadosi thing that distinguishes language from other cognition. However, et al., 2012), but the PDC can offer something more to the extent linguistic utterances clearly differ from other actions in that that it draws on the mechanisms of memory retrieval, attention, they have both a goal (e.g., to communicate) and a mean- serial order maintenance, and motor planning in understanding ing, while complex actions have a goal (e.g., to make coffee), what is more vs. less efficient. Similarly, Bybee (2006) and oth- and a hierarchical plan to realize the goal, but no inherent ers make important claims that language use, broadly construed, meaning. This meaning and its interplay with utterance form, underlies language typology and change, but the PDC aims to meted out over time as the language is planned, produced, be more specific: Language producibility, more than learnabil- and comprehended, would seem to be a critical aspect of what ity or comprehensibility, drives language form. The reasons for makes language unlike non-linguistic cognition. Again, work this claim again invoke mechanistic accounts of language pro- toward a mechanistic account of how language is planned and duction to explain what is difficult, how producers manage that uttered may have consequences well beyond the field of language difficulty, and how they are the primary controllers of utter- production. ance form. There’s a great deal of work remaining in order to realize this goal of a mechanistic account of language pro- ACKNOWLEDGMENTS duction, including extensions beyond the lexico-syntactic focus Preparation of this article was supported by grants from the of this article. Working toward a more mechanistic account National Institutes of Health (R01 HD047425), the National is important because links to memory, action planning, and Science Foundation (BCS 1123788), and the Wisconsin Alumni other non-linguistic domains can ground the PDC approach Research Fund. This work greatly benefitted from comments in broader cognitive processes and avoid potential circulari- by Gary Lupyan, Jenny Saffran, Mark Seidenberg, and mem- ties among what is efficient, common, easy, salient, and other bers of the Language and Cognitive Neuroscience Lab at constructs that are invoked in many accounts of language form the University of Wisconsin-Madison. Errors are of course and use. my own. www.frontiersin.org April 2013| Volume4| Article226 | 13 MacDonald Language production shapes form and comprehension Bresnan, J., and Ford, M. (2010). Cooper, R., and Shallice, T. (2000). Ferreira, V. S., and Dell, G. S. (2000). REFERENCES Acheson, D. J., and MacDonald, Predicting syntax: processing dative Contention scheduling and the Effect of ambiguity and lexical M. C. (2009). Verbal working constructions in American and control of routine activities. Cogn. availability on syntactic and lexi- memory and language produc- Australian varieties of English. Neuropsychol. 17, 297–338. cal production. Cognit. Psychol. 40, tion: common approaches to Language 86, 168–213. Cowan, N. (2005). Working Memory 296–340. the serial ordering of verbal Bybee, J. (2006). From usage to gram- Capacity. New York, NY: Taylor and Ferrera, V. P. (2000). Task-dependent information. Psychol. Bull. 135, mar: the mind’s response to repeti- Francis. modulation of the sensorimotor 50–68. tion. Language 82, 711–733. Culbertson, J., Smolensky, P., and transformation for smooth pursuit Allen, J., and Seidenberg, M. S. (1999). Bybee, J., and McClelland, J. L. (2005). Legendre, G. (2012). Learning eye movements. J. Neurophysiol. 84, “The emergence of grammaticality Alternatives to the combinatorial biases predict a word order 2725–2738. in connectionist networks,” in paradigm of linguistic theory based universal. Cognition 122, 306–329. Fidelholtz, J. (1975). Word frequency The Emergence of Language, ed on domain general principles of Dahan, D., Tanenhaus, M. K., and and vowel reduction in English. B. MacWhinney (Mahwah, NJ: human cognition. Linguist. Rev. 22, Chambers, C. G. (2002). Accent Chic. Linguist. Soc. 11, 200–213. Erlbaum), 115–151. 381–410. and reference resolution in spoken- Fodor, J. D., and Inoue, A. (1994). The Allen, K., Ibara, S., Seymour, A., Carreiras, M., Duñabeitia, J. A., language comprehension. J. Mem. diagnosis and cure of garden paths. Cordova, N., and Botvinick, M. Vergara, M., de la Cruz-Pavía, I., Lang. 47, 292–314. J. Psycholinguist. Res. 23, 407–434. (2010). Abstract structural repre- and Laka, I. (2010). Subject relative Dakin, S.C., Mareschal, I., andBex,P. Fox Tree, J. E., and Clark, H. H. (1997). sentations of goal-directed behavior. clauses are not universally easier J. (2005). An oblique effect for local Pronouncing “the” as “thee” to sig- Psychol. Sci. 21, 1518–1524. to process: evidence from Basque. motion: psychophysics and natural nal problems in speaking. Cognition Altmann, G. T. M.,van Nice,K.Y., Cognition 115, 79–92. movie statistics. J. Vis. 5, 878–887. 62, 151–167. Garnham, A., and Henstra, J.-A. Chafe, W. (1976). “Givenness, con- Deco, G., and Rolls, E. T. (2005). Frazier, L. (1987). “Sentence process- (1998). Late closure in context. trastiveness, definiteness, subjects, Sequential memory: a putative ing: a tutorial review,” in Attention J. Mem. Lang. 38, 459–484. topics, and point of view,” in neural and synaptic dynamical and Performance, Vol. 12. The psy- Anderson, J. R. (1983). Retrieval of Subject and Topic, ed C. Li mechanism. J. Cogn. Neurosci. 17, chology of reading, ed M. Coltheart information from long-term mem- (New York, NY: Academic Press), 294–307. (Hove: Erlbaum). ory. Science 220, 25–30. 25–55. Dell, G. S., Burger, L. K., and Svec, Frazier, L. (1995). Constraint satisfac- Aslin, R. N., and Newport, E. L. (2012). Chang, F. (2009). Learning to order W. R. (1997). Language production tion as a theory of sentence pro- Statistical learning: from acquiring words: a connectionist model of and serial order: a functional anal- cessing. J. Psycholinguist. Res. 24, specific items to forming general heavy NP shift and accessibility ysis and a model. Psychol. Rev. 104, 437–468. rules. Curr.Dir.Psychol.Sci. 21, effects in Japanese and English. 123–147. Gennari,S.P., and MacDonald,M. C. 170–176. J. Mem. Lang. 61, 374–397. Dikker, S., Rabagliati, H., Farmer, T. (2008). Semantic indeterminacy in Babyonyshev, M., and Gibson, E. Chang, F., Dell, G. S., and Bock, K. A., and Pylkkänen, L. (2010). Early object relative clauses. J. Mem. Lang. (1999). The complexity of nested (2006). Becoming syntactic. Psychol. occipital sensitivity to syntactic cat- 58, 161–187. structures in Japanese. Language 75, Rev. 113, 234–272. egory is based on form typicality. Gennari, S. P., and MacDonald, M. 423–450. Chomsky, N. (1957). Syntactic Psychol. Sci. 21, 629–634. C. (2009). Linking production and Bard, E. G., Anderson, A. H., Chen, Structures. The Hague: Mouton. Duran, N. D., Dale, R., and Kreuz, comprehension processes: the case Y.,Nicholson, H. B. M.,Havard, C., Chomsky, N., and Miller, G. A. (1963). R. J. (2011). Listeners invest in an of relative clauses. Cognition 111, and Dalzel-Job, S. (2007). Let’s you “Introduction to the formal analysis assumed other’s perspective despite 1–23. do that: sharing the cognitive bur- of natural languages,” in Handbook cognitive cost. Cognition 121, 22–40. Gennari,S.P., Mirkovic, ´ J., and dens of dialogue. J. Mem. Lang. 57, of Mathematical Psychology,Vol.2, Echols, C. H., and Newport, E. L. MacDonald, M. C. (2012). Animacy 616–641. eds R. D. Luce, R. R. Bush, and (1992). The role of stress and posi- and competition in relative clause Bock, J. K. (1982). Toward a cogni- E. Galanter (New York, NY: Wiley), tion in determining first words. production: a cross-linguistic tive psychology of syntax: infor- 269–321. Lang. Acquis. 2, 189–220. investigation. Cognit. Psychol. 65, mation processing contributions to Christiansen,M.H., andChater, Evans, N., and Levinson, S. C. (2009). 141–176. sentence formulation. Psychol. Rev. N. (2001). “Finite models of The myth of language universals: Gibson, B. M., Wasserman, E. A., 89, 1–47. infinite language: a connection- language diversity and its impor- and Kamil, A. C. (2007). Pigeons Bock, J. K., and Warren, R. K. (1985). ist approach to recursion,” in tance for cognitive science. Behav. and people select efficient routes Conceptual accessibility and syntac- Connectionist Psycholinguistics, eds Brain Sci. 32, 429–448. when solving a one-way “trav- tic structure in sentence formula- M. H. Christiansen and N. Chater Feist, M. I. (2010). “Inside in and eling salesperson” task. J. Exp. tion. Cognition 21, 47–67. (Westport, CT: Ablex), 138–176. on: typological and psycholin- Psychol. Anim. Behav. Process. 33, Botvinick, M., and Plaut, D. C. (2004). Christianson, K., and Ferreira, F. guistic perspectives,” in Language, 244–261. Doing without schema hierarchies: (2005). Conceptual accessibility and Cognition, and Space, eds V. Evans Gibson, E. (1998). Linguistic complex- a recurrent connectionist approach sentence production in a free word and P. Chilton (London: Equinox). ity: locality of syntactic dependen- to normal and impaired routine order language (Odawa). Cognition Ferreira, F. (1991). Effects of length cies. Cognition 68, 1–76. sequential action. Psychol. Rev. 111, 98, 105–135. and syntactic complexity on initia- Gibson, E., Pearlmutter, N., Canseco- 395–429. Clahsen, H., Lück, M., and Hahne, tion times for prepared utterances. Gonzalez, E., and Hickok, G. Branigan, H. P., Pickering, M. J., and A. (2007). How children process J. Mem. Lang. 30, 210–233. (1996). Recency preference in Cleland, A. A. (1999). Syntactic over-regularizations: evidence from Ferreira, F. (1994). Choice of passive the human sentence processing priming in written production: evi- event-related brain potentials. voice is affected by verb type and mechanism. Cognition 59, 23–59. dence for rapid decay. Psychon. Bull. J. Child Lang. 34, 601–622. animacy. J. Mem. Lang. 33, 715–736. Givón, T. (1985). “Function, struc- Rev. 6, 635–640. Connine,C. M., andClifton,C. J. Ferreira, F., and Swets, B. (2002). How ture, and language acquisition,” Bresnan, J., Cueni, A., Nikitina, T., and (1987). Interactive use of lexi- incremental is language production? in The Crosslinguistic Study of Baayen, R. H. (2007). “Predicting cal information in speech percep- Evidence from the production of Language Acquisition,Vol.2,ed D. the dative alternation,” in Cognitive tion. J. Exp. Psychol. Hum. Percept. utterances requiring the computa- Slobin. (Hillsdale, NJ: Erlbaum), Foundations of Interpretation, Perform. 13, 291–299. tion of arithmetic sums. J. Mem. 1005–1028. eds G. Boume, I. Kraemer, and Conrad, R., and Hull, A. J. (1964). Lang. 46, 57–84. Goldberg, A. (2006). Constructions at J. Zwarts (Amsterdam: Royal Information, acoustic confusion Ferreira, V. S., and Bock, K. (2006). Work: The Nature of Generalization Netherlands Academy of Science), and memory span. Br.J.Psychol. 55, The functions of structural priming. in Language. Oxford: Oxford 69–94. 429–432. Lang. Cogn. Process. 21, 1011–1029. University Press. Frontiers in Psychology | Language Sciences April 2013 | Volume 4 | Article 226 | 14 MacDonald Language production shapes form and comprehension Grodner, D., and Gibson, E. (2005). (2011). Tracking talking: dual task sentence comprehension,” in visually situated sentence pro- Consequences of the serial nature of costs of planning and produc- Handbook of Psycholinguistics, eds duction. Front. Psychol. 2:396. doi: linguistic input for sentenial com- ing speech for young versus older M. Traxler and M. A. Gernsbacher 10.3389/fpsyg.2011.00396 plexity. Cogn. Sci. 29, 261–290. adults. Aging Neuropsychol. Cogn. (New York, NY: Academic Press), Newmeyer, F. J. (1998). Language Form Grossberg, S. (1978). A theory of 18, 257–279. 581–611. and Language Function. Cambridge, human memory: self-organization Kimball, J. (1973). Seven principles of MacDonald, M. C., and Thornton, R. MA: MIT Press. and performance of sensory-motor surface structure parsing in natural (2009). When language comprehen- O’Brien, J. L., and Raymond, J. E. codes, maps, and plans. Prog. Theor. language. Cognition 2, 15–47. sion reflects production constraints: (2012). Learned predictiveness Biol. 5, 233–374. Kraljic, T., Samuel, A. G., and Brennan, resolving ambiguities with the help speeds visual processing. Psychol. Hale, J. (2006). Uncertainty about the S. E. (2008). First impressions and of past experience. Mem. Cognit. 37, Sci. 23, 359–363. rest of the sentence. Cogn. Sci. 30, last resorts: how listeners adjust to 1177–1186. O’Grady, W. (2011). “Relative clauses: 643. speaker variability. Psychol. Sci. 19, Mack,J.E., Clifton, C., Frazier,L., and processing and acquisition,” in Hart, B., and Risley, T. R. (1995). 332–338. Taylor, P. V. (2012). (Not) Hearing The Acquisition of Relative Clauses: Meaningful Differences in the Kuchinsky, S. E., Bock, K., and Irwin, optional subjects: the effects of Processing, Typology and Function, Everyday Experience of Young D. E. (2011). Reversing the hands pragmatic usage preferences. ed E. Kidd (Amsterdam: John American Children.Baltimore, MD: of time: changing the mapping from J. Mem. Lang. 67, 211–223. Benjamins), 13–38. Brookes. seeing to saying. J. Exp. Psychol. MacNeilage, P.F., and Davis,B.L. Piantadosi, S. T., Tily, H., and Gibson, Hartley, T., and Houghton, G. (1996). A Learn. Mem. Cogn. 37, 748–756. (2000). On the origin of internal E. (2012). The communicative func- linguistically constrained model of Kuperman, V., and Bresnan, J. (2012). structure of word forms. Science tion of ambiguity in language. short-term memory for nonwords. The effects of construction prob- 288, 527–531. Cognition 122, 280–291. J. Mem. Lang. 35, 1–31. ability on word durations dur- Miller, G. A. (1956). The magical num- Pickering, M. J., and Garrod, S. (2007). Hawkins, J. A. (2004). Efficiency and ing spontaneous incremental sen- ber seven, plus or minus two: some Do people use language production Complexity in Grammars. Oxford: tence production. J. Mem. Lang. 66, limits on our capacity for process- to make predictions during com- Oxford University Press. 588–611. ing information. Psychol. Rev. 63, prehension? Trends Cogn. Sci. 11, Horton, W. S., and Keysar, B. (1996). Lashley, K. S. (1951). “The problem of 81–97. 105–110. When do speakers take into account serial order in behavior,” in Cerebral Miller, G. A., and Chomsky, N. (1963). Postle, B. R. (2006). Working memory common ground? Cognition 59, Mechanisms in Behavior; the Hixon “Finitary models of language users,” as an emergent property of the mind 91–117. Symposium, ed L. A. Jeffress (Oxford in Handbook of Mathematical and brain. Neuroscience 139, 23–38. Hudson, R. (1996). The difficulty England: Wiley), 112–146. Psychology,Vol.2, eds R. D. Luce, Ramscar, M., and Yarlett, D. (2007). of (so-called) self-embedded Levelt, W. J. M. (1982). “Linearization R. R. Bush, and E. Galanter. (New Linguistic self-correction in the structures. Work.Pap.Linguist. 8, in describing spatial networks,” in York, NY: Wiley), 419–491. absence of feedback: a new 283–314. Processes, Beliefs, and Questions, Miller, G. A., and Selfridge, J. A. (1950). approach to the logical prob- Hudson Kam, C. L., and Newport, E. eds S. Peters and E. Saarinen Verbal context and the recall of lem of language acquisition. Cogn. L. (2009). Getting it right by getting (Dordrecht: Reidel), 199–220. meaningful material. Am. J. Psychol. Sci. 31, 927–960. it wrong: when learners change lan- Levy, R. (2008). Expectation-based 63, 176–185. Reali, F., and Christiansen, M. H. guages. Cognit. Psychol. 59, 30–66. syntactic comprehension. Cognition Mitchell, D. C., and Brysbaert, M. (2007). Word chunk frequencies Jackendoff, R. (2002). Foundations 106, 1126–1177. (1998). “Challenges to recent the- affect the processing of pronomi- of Language: Brain, Meaning, Lightfoot, D. (1999). The Development ories of cross-linguistic differences nal object-relative clauses. Q. J. Exp. Grammar, Evolution. Oxford: of Language: Acquisition, Change, in parsing: evidence from Dutch,” Psychol. 60, 161–170. Oxford University Press. and Evolution. Oxford: Blackwell. in Sentence Processing: A Cross- Rokem, A., and Silver, M. A. (2009). Jackendoff, R. (2007). Linguistics in Lin, C. J. C. (2008). The process- Linguistic Perspective, ed D. Hillert A model of encoding and decod- cognitive science: the state of the art, ing foundation of head-final relative (San Diego, CA: Academic Press), ing in V1 and MT accounts for Linguist. Rev. 24, 347–401. clauses. Lang. Linguist. 9, 813–838. 313–335. motion perception anisotropies in Jaeger, T. F., and Tily, H. (2011). On lan- MacDonald, M. C. (1994). Probabilistic Miyata, H., and Fujita, K. (2011). the human visual system. Brain Res. guage “utility”: processing complex- constraints and syntactic ambiguity Flexible route selection by pigeons 1299, 3–16. ity and communicative efficiency. resolution. Lang. Cogn. Process. 9, (Columba livia)onacomputer- Rosenbaum, D. A., Cohen, R. G., Jax, S. WIREs Cogn. Sci. 2, 323–335. 157–201. ized multi-goal navigation task with A.,Weiss, D. J.,and vander Wel, R. Jäger, G., and Rosenbach, A. (2008). MacDonald, M. C. (1999). and without an “obstacle.” J. Comp. (2007). The problem of serial order Priming and unidirectional lan- “Distributional information in Psychol. 125, 431–435. in behavior: Lashley’s legacy. Hum. guage change. Theor. Linguist. 34, language comprehension, produc- Montag, J. L., and MacDonald, M. Mov. Sci. 26, 525–554. 85–113. tion, and acquisition: three puzzles C. (2009). “Word order doesn’t Roth, F. P. (1984). Accelerating lan- Jin, D. Z. (2009). Generating vari- and a moral,” in The Emergence matter: relative clause produc- guage learning in young children. able birdsong syllable sequences of Language, ed B. MacWhinney tion in English and Japanese,” in J. Child Lang. 11, 89–107. with branching chain networks (Mahwah, NJ: Lawrence Erlbaum Proceedings of the 31th Annual Rumelhart, D. E., and McClelland, J. in avian premotor nucleus HVC. Associates Publishers), 177–196. Conference of the Cognitive Science L. (1986). “On learning the past Phys. Rev. E. Stat. Nonlin. Soft MacDonald, M. C., and Christiansen, Society, eds N. A. Taatgen and H. tense of English verbs,” in Parallel Matter Phys. 80(5 Pt 1):051902. doi: M. H. (2002). Reassessing work- van Rijn (Austin, TX: Cognitive Distributed Processing: Explorations 10.1103/PhysRevE.80.051902 ing memory: comment on Just and Science Society), 2594–2599. in the Microstructure of Cognition, Kaiser, E. (2012). Taking action: Carpenter (1992) and Waters and Müller, H., and Sternad, D. (2004). Vol. 2, eds J. L. McClelland, a cross-modal investigation of Caplan (1996). Psychol. Rev. 109, Decomposition of variability in D. E. Rumelhart and the PDP discourse-level representations. 35–54. the execution of goal-oriented Research Group. (Cambridge, Front. Psychol. 3:156. doi: 10.3389/ MacDonald, M. C., Pearlmutter, N. tasks: three components of skill MA: MIT Press/Bradford Books), fpsyg.2012.00156 J., and Seidenberg, M. S. (1994). improvement. J. Exp. Psychol. 217–271. Karpicke, J. D. (2012). Retrieval-based The lexical nature of syntactic Hum. Percept. Perform. 30, Scott-Phillips, T. C., and Kirby, S. learning: active retrieval promotes ambiguity resolution. Psychol. Rev. 212–233. (2010). Language evolution in the meaningful learning. Curr. Dir. 101, 676–703. Myachykov, A., Thompson, D., laboratory. Trends Cogn. Sci. 14, Psychol. Sci. 21, 157–163. MacDonald, M. C., and Seidenberg, Garrod, S., and Scheepers, C. 411–417. Kemper, S., Hoffman, L., Schmalzried, M. S. (2006). “Constraint satis- (2011). Referential and visual Seidenberg, M. S. (1997). Language R., Herman, R., and Kieweg, D. faction accounts of lexical and cues to structural choice in acquisition and use: learning and www.frontiersin.org April 2013| Volume4| Article226 | 15 MacDonald Language production shapes form and comprehension applying probabilistic constraints. first-pass structural attachment abandoned context words affect a role for (co-)articulatory fluency? Science 275, 1599–1604. preferences with dynamic speech production. Acta Psychol. J. Mem. Lang. 58, 48–65. Seidenberg, M. S., and Zevin, J. D. grammars and recursive neural 140, 218–229. Yamashita, H., and Chang, F. (2001). (2006). “Connectionist models in networks. Cognition 88, 133–169. Van Petten, C., and Luka, B. J. (2012). “Long before short” preference developmental cognitive neuro- Tanaka, M. N., Branigan, H. P., Prediction during language compre- in the production of a head- science: critical periods and the McLean, J. F., and Pickering, hension: benefits, costs, and ERP final language. Cognition 81, paradox of success,” in Attention M. J. (2011). Conceptual influ- components. Int. J. Psychophysiol. B45–B55. and Performance XXI: Processes ences on word order and voice 83, 176–190. of Change in Brain and Cognitive in sentence production: evidence Wanner, E., and Maratsos, M. (1978). Conflict of Interest Statement: The Development, eds Y. Munakata from Japanese. J. Mem. Lang. 65, “An ATN approach to compre- authors declare that the research and M. Johnson (Oxford: Oxford 318–330. hension,” in Linguistic Theory was conducted in the absence of any University Press), 585–612. Tarr, M. J., and Gauthier, I. (2000). and Psychological Reality, eds M. commercial or financial relationships Singleton, J. L., and Newport, E. L. FFA: a flexible fusiform area for Halle, J. Bresnan, and G. A. Miller. that could be construed as a potential (2004). When learners surpass subordinate-level visual processing (Cambridge, MA US: MIT Press), conflict of interest. their models: The acquisition of automatized by expertise. Nat. 119–161. American Sign Language from Neurosci. 3, 764–769. Warren, R. M., and Sherman, G. L. Received: 08 November 2012; paper inconsistent input. Cognit. Psychol. Thornton, R., MacDonald, M. C., and (1974). Phonemic restorations pending published: 25 November 2012; 49, 370–407. Gil, M. (1999). Pragmatic con- based on subsequent context. accepted: 11 April 2013; published Smith, M., and Wheeldon, L. (2004). straint on the interpretation of com- Percept. Psychophys. 16, 150–156. online: 26 April 2013. Horizontal information flow in spo- plex noun phrases in Spanish and Wasow, T. (1997). Remarks on gram- Citation: MacDonald MC (2013) How ken sentence production. J. Exp. English. J. Exp. Psychol. Learn. Mem. matical weight. Lang. Variation language production shapes language Psychol. Learn. Mem. Cogn. 30, Cogn. 25, 1347–1365. Change 9, 81–105. form and comprehension. Front. Psychol. 675–686. Traxler, M. J., Morris, R. K., and Seely, Weiner, E. J., and Labov, W. (1983). 4:226. doi: 10.3389/fpsyg.2013.00226 Stallings, L. M., MacDonald, M. C., R. E. (2002). Processing subject Constraints on the agentless passive. This article was submitted to Frontiers and O’Seaghdha, P. G. (1998). and object relative clauses: evidence J. Linguist. 19, 29–58. in Language Sciences, a specialty of Phrasal ordering constraints in from eye movements. J. Mem. Lang. Wells, J. B., Christiansen, M. H., Frontiers in Psychology. sentence production: Phrase 47, 69–90. Race, D. S., Acheson, D. J., Copyright © 2013 MacDonald. This is length and verb disposition in Tse, C.-S., Li, Y., and Altarriba, J. and MacDonald, M. C. (2009). an open-access article distributed under heavy-NP shift. J. Mem. Lang. 39, (2011). The effect of semantic relat- Experience and sentence process- the terms of the Creative Commons 392–417. edness on immediate serial recall ing: statistical learning and relative Attribution License,which permits use, Stemberger, J. P. (1989). Speech errors and serial recognition. Q. J. Exp. clause comprehension. Cognit. distribution and reproduction in other in early child language production. Psychol. 64, 2425–2437. Psychol. 58, 250–271. forums, provided the original authors J. Mem. Lang. 28, 164–188. Tydgat, I., Diependaele, K., Hartsuiker, Woodward, A. J., Macken, W. J., and and source are credited and subject to any Sturt, P., Costa, F., Lombardo, V., R. J., and Pickering, M. J. (2012). Jones, D. M. (2008). Linguistic copyright notices concerning any third- and Frasconi, P. (2003). Learning How lingering representations of familiarity in short-term memory: party graphics etc. Frontiers in Psychology | Language Sciences April 2013 | Volume 4 | Article 226 | 16 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Frontiers in Psychology Pubmed Central

How language production shapes language form and comprehension

Frontiers in Psychology , Volume 4 – Apr 26, 2013

Loading next page...
 
/lp/pubmed-central/how-language-production-shapes-language-form-and-comprehension-Su2W9BUPwr

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Pubmed Central
Copyright
Copyright © 2013 MacDonald.
ISSN
1664-1078
eISSN
1664-1078
DOI
10.3389/fpsyg.2013.00226
Publisher site
See Article on Publisher Site

Abstract

HYPOTHESIS AND THEORY ARTICLE published: 26 April 2013 doi: 10.3389/fpsyg.2013.00226 How language production shapes language form and comprehension Maryellen C. MacDonald* Department of Psychology, University of Wisconsin-Madison, Madison, WI, USA Edited by: Language production processes can provide insight into how language comprehension Charles Clifton, University of works and language typology—why languages tend to have certain characteristics more Massachusetts Amherst, USA often than others. Drawing on work in memory retrieval, motor planning, and serial Reviewed by: order in action planning, the Production-Distribution-Comprehension (PDC) account links Gary Dell, University of Illinois at work in the fields of language production, typology, and comprehension: (1) faced Urbana-Champaign, USA Fernanda Ferreira, University of with substantial computational burdens of planning and producing utterances, language South Carolina, USA producers implicitly follow three biases in utterance planning that promote word order Joan Bresnan, Stanford University, choices that reduce these burdens, thereby improving production fluency. (2) These USA choices, repeated over many utterances and individuals, shape the distributions of *Correspondence: utterance forms in language. The claim that language form stems in large degree from Maryellen C. MacDonald, Department of Psychology, producers’ attempts to mitigate utterance planning difficulty is contrasted with alternative University of Wisconsin-Madison, accounts in which form is driven by language use more broadly, language acquisition 1202 West Johnson St., Madison, processes, or producers’ attempts to create language forms that are easily understood WI 53706, USA. by comprehenders. (3) Language perceivers implicitly learn the statistical regularities e-mail: mcmacdonald@wisc.edu in their linguistic input, and they use this prior experience to guide comprehension of subsequent language. In particular, they learn to predict the sequential structure of linguistic signals, based on the statistics of previously-encountered input. Thus, key aspects of comprehension behavior are tied to lexico-syntactic statistics in the language, which in turn derive from utterance planning biases promoting production of comparatively easy utterance forms over more difficult ones. This approach contrasts with classic theories in which comprehension behaviors are attributed to innate design features of the language comprehension system and associated working memory. The PDC instead links basic features of comprehension to a different source: production processes that shape language form. Keywords: language acquisition, motor control, language production, serial order, language comprehension, syntax, language typology, working memory a central role for experience in development and in adult perfor- INTRODUCTION Humans are capable of a remarkable number of highly com- mance. plex behaviors—we plan ahead, remember the past, reason, The statistical properties of the input have a similarly cru- infer, and invent. The origins of intelligent behavior are at cial role in some accounts of language use, including the role the core of classic debates in cognitive science concerning the of linguistic experience in acquisition (Hart and Risley, 1995) contributions of innate capacities and experience in the devel- and inadult comprehensionprocesses(MacDonald et al., 1994). opment of thought, perception, and action. For example, the However, the nature of the argument is critically different in fact that perception of motion in cardinal directions (verti- vision and in language. Visual experience reflects the nature of cal, horizontal) is superior to that in oblique directions has the physical world: Vision scientists do not need to explain why been attributed to the greater number of cells in visual cortex gravity creates many experiences of downward motion, and no devoted to processing cardinal motion directions than oblique one expects face perception researchers to explain why faces have ones (Rokem and Silver, 2009), and this result in turn is particular shapes. In language, however, the input to the perceiver thought to arise from visual experience: There are more motion is itself the consequence of language behavior—it is the utterances events in the world in cardinal directions than in oblique ones produced by other language users, who have their own cognitive (Dakin et al., 2005). Similarly, experience-based accounts of systems presumably shaped by their own experiences. This situ- face perception hold that face recognition behavior diverges ation lends potential circularity to experience-based accounts of from object recognition because perceivers’ visual experience language (Frazier, 1995), requiring solutions for two unknowns at with faces differs in critical ways from their experience with once: as in vision, language researchers must develop an account objects (Tarr and Gauthier, 2000). While such accounts don’t of the effects of experience on perception, but unlike in vision, deny innate factors in perception, they are notable in ascribing language researchers must also consider why the experience—the www.frontiersin.org April 2013 | Volume 4 | Article 226 | 1 MacDonald Language production shapes form and comprehension language—has the character it does. This difficult task is com- development and organization of plans for output sequences as pounded by the fact that the psycholinguists who study language “both the most important and also the most neglected prob- use are typically not the same people as the linguists who study the lem in cerebral physiology” (p. 114). He argued that complex nature of language form, so that there is a gulf between linguistic sequential actions such as speaking must be guided by a plan that theories of the nature of language and psycholinguists’ accounts is developed before execution, a view that continues to pervade of the effects of experience with language patterns. research in motor behavior, including language production. The This article is a step toward bridging this divide, offering construction of motor plans is a cognitively demanding activ- insight into both the origin of language form and also the effect ity; developing the utterance plan can be more demanding than of experience with these forms. The Production-Distribution- speaking itself (Kemper et al., 2011). The significant computa- Comprehension (PDC) account, first sketched in MacDonald tional difficulty of constructing and maintaining an utterance (1999) andelaboratedin workdescribed here, holds that the plan is a key component of the PDC, and so we consider these memory and planning demands of language production strongly planning operations in some detail. affect the form of producers’ utterances. Constraints imposed by the production process have two important consequences. DEVELOPMENT AND CARE OF THE UTTERANCE PLAN First, they contribute to understanding regularities in linguistic Language planning shares features of both high-level non- form: why languages exhibit particular properties, with differ- linguistic action planning and more fine grained motor control. ent frequencies across languages. Second, they determine many In high-level action plans, some elements have only loosely con- aspects of language comprehension. The claim is not that all strained sequences. In making coffee, an example extensively aspects of language form and comprehension can be traced to the discussed inresearchonactionplanning and control (Cooper computational demands of language production, but rather that and Shallice, 2000; Botvinick and Plaut, 2004), the coffee, cream, production’s impact in these areas is so pervasive that understand- and sugar can go into the cup in any order. Similarly, in some ing production becomes essential to explaining why language is (though by no means all) aspects of language planning, some the way it is, and why language comprehension works the way elements may be ordered in several ways, as in Jane bought a ham- it does. mer and some batteries at the hardware store, vs. At the hardware In this article I describe the Production, Distribution, and store, Jane bought some batteries and a hammer. Other aspects of Comprehension components of the PDC in that order, focus- action/motor plans are far more constrained—one must move ing particularly on lexico-syntactic phenomena. The section the hand to the coffee cup before grasping it, and in the case of entitled The First Step in the PDC: Production Difficulty and its language planning, there are language-specific constraints limit- Amelioration reviews the memory and control demands of lan- ing the range of permissible word orders, for example excluding guage production, producers’ attempts to mitigate them, and the hardware at store the. Thus language producers have word order patterns of word order, sentence form, and lexical-sentence pair- options in some cases but not others, and when there are options, ings that result. Findings in motor control, memory retrieval, producers must very rapidly settle on one form and inhibit oth- and short term maintenance suggest that many properties of ers from interfering, so as not to make speech error blends of language production that affect utterance form also arise in alternative forms such as some hammer and a batteries.This action and motor planning more generally. Next, the section behavior is an example of a winner-take-all process, and winner- entitled Distributional Regularities and Language Typology con- take-all neural mechanisms form an important part of accounts siders the effects of language production on language form and and computational models of both language production (Hartley views the potential contributions of the PDC in the context of and Houghton, 1996; Dell et al., 1997) and non-linguistic motor other accounts of why languages have some properties more behavior, including visual search (Ferrera, 2000) and the “syn- than others. Finally, ComprehensionConsequences inthe PDC tax” of birdsong (Jin, 2009). This winner-take-all property of addresses comprehension, showing that the PDC provides a dif- language production is critical in accounts of how producers acti- ferent framework for thinking about sentence comprehension vate the correct serial order of elements in articulation (Hartley and offers a different explanation of some classic results. and Houghton, 1996), and it provides our first example of how properties of motor planning affect distributional patterns in the THE FIRST STEP IN THE PDC: PRODUCTION DIFFICULTY language, in that this property affects the incidence of speech AND ITS AMELIORATION errors. Language production is a highly complex motor behavior, requir- The developing utterance plan must be maintained in an exe- ing the translation of conceptual information into an intricate cutable state as it is being developed. The plan is effectively “the sequence of motor commands to allow speaking, signing, writing, memory for what is to come” (Rosenbaum et al., 2007, p. 528), or typing. Although “production difficulty” and “motor con- with all the maintenance burdens of other short-term memories. trol” might suggest a discussion of articulation, here we consider Indeed, verbal working memory studies offer important insights difficulty arising in the development of the plan for the utter- into some of the memory demands of language production. In ance, well ahead of articulation . Lashley (1951) considered the both serial recall tasks (in which unrelated words are recalled in the same order they were presented) and language produc- 1 tion tasks (such as describing pictures), elements in the utterance “Utterance” here refers to all modalities (speaking, writing, signing). Each plan tend to interfere with one another, affecting the fluency of modality’s unique production demands should influence the distribution of forms in that modality, but those effects aren’t discussed here. speech. For example, phonological overlap among elements in Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 2 MacDonald Language production shapes form and comprehension the utterance plan increases the difficulty in both production and which include developing the plan, maintaining it, monitoring memory tasks (Acheson and MacDonald, 2009), and semantic the state of execution, and shifting attentional focus as the plan overlap between words increases errors in language production is executed over time. Second, they illustrate how speakers learn (Smith and Wheeldon, 2004) and memory tasks (Tse et al., 2011). implicit strategies to mitigate production difficulty, in this case Conversely, production of the correct serial order of elements learning to allocate more attention to the upcoming plan as they is improved by increased linguistic frequency or coarticulatory become more fluent, and learning to favor early execution and experience, both for memory tasks (Woodward et al., 2008)and incremental planning, with delaying tactics and additional dam- language production (Dell et al., 1997). Thus, production plan- age control if the plan runs out. And third, production-related ning has inherent working memory demands, with consequent learning affects the distribution of utterance forms that people interference and other pitfalls well known to memory researchers. produce, in this case the rate and distribution of speech errors and Because planning precedes execution, a key question in lan- pauses in utterances. The intersection of these last two points— guage production concerns the degree of advance planning before that the computational demands of language production can be execution begins. Language production is said to be incremen- mitigated, but with consequences for utterance form—will reap- tal, meaning that partial planning, execution, and subsequent pear below as a force in the distribution of syntactic forms in planning are interleaved. The scope of advance planning varies languages. in different circumstances and is at least partially under the MINIMIZING DIFFICULTY DURING PRODUCTION producer’s strategic control (Ferreira and Swets, 2002). Again, Incremental production—the interleaving of plan and production behavior is shaped by learned implicit strategies that execution—works only if new plan segments can be developed maximize fluency, as the scope of planning strikes a balance between competing demands. On the one hand, initiating exe- at a rate that keeps up with execution. New plan development in turn relies on retrieval from long term memory, and when cution before much planning is complete allows producers to begin speaking earlier, avoiding long pauses and retaining the this retrieval fails or requires extra time, production is delayed or derailed. We next review three memory-related production floor in a conversation. Early execution also avoids the mem- ory burden of maintaining and executing a large plan, as more biases that have substantial consequences for lexico-syntactic distributions in utterance form. complex plans require more time to initiate execution, both in speech (Ferreira, 1991) and in non-linguistic motor behaviors Easy First: a source of word order flexibility (Rosenbaum et al., 2007). However, interleaving planning and execution has the occasional negative consequence of the pro- As anyone who has been in a tip-of-the-tongue state knows, ducer finishing the executable portion of the plan before the next some words are more easily retrieved from memory than oth- portion is ready. Rather than letting everything grind to a halt, ers. This fact has enormous influence on language form, because speakers in this situation attempt to gain extra planning time by easily retrieved words and phrases tend to appear both earlier lengthening words or adding optional words and pauses, yielding in utterances and at more prominent syntactic positions (e.g., utterances such as “Have you seen theee ...um ...?” (Fox Tree sentence subject) than ones that are more difficult to retrieve and Clark, 1997; Ferreira and Dell, 2000). (Bock, 1982; Tanaka et al., 2011). An Easy First bias in incre- Beyond juggling planning and executing, language produc- mental production allows execution of utterances to begin early, ers must also keep track of where they are in the plan as it is starting with easily planned elements, leaving more time for being executed. Tracking the state of progress through the plan planning of more difficult ones. “Easier” (also termed more acces- is critical for avoiding repetitions, omissions and other sequenc- sible or available) words and phrases have been described as ing errors, but it comes at a cost, in that tracking plan progress more frequent, shorter (both number of words in a phrase, and itself carries substantial additional attention or maintenance bur- number of syllables in a word), less syntactically complex, more dens (Botvinick and Plaut, 2004). At the same time, the memory important or conceptually salient to the speaker, and previously for what has been uttered cannot remain too strong, because mentioned (“given”) in the discourse (Levelt, 1982; Bock and recently-executed actions can interfere with upcoming ones, lead- Warren, 1985; Tanaka et al., 2011). There are enough different ing to perseverations and other errors (Tydgat et al., 2012). The forces affecting ease of planning that the claims can seem cir- speaker must therefore balance the various subtasks in utterance cular: Easy entities are easy because they appear earlier in the planning in order to “activate the present, deactivate the past, and utterance. However, the essential claim—that utterance planning prepare to activate the future” (Dell et al., 1997, p. 123; a non- difficulty affects speakers’ choices of word order and sentence linguistic example is Deco and Rolls, 2005). An efficient allocation structure—gains external validity in several ways. First, difficulty of attention to past, present, and future is learned over time: stems from ease of retrieval from long term memory, and many Fluent adult speech reflects a bias toward the future, with compar- of the factors that promote early positioning in an utterance plan atively more anticipation errors (elements of the upcoming plan also predict the early positioning and accuracy of word recall incorrectly influencing the current execution) than perseverations in verbal memory tasks, including word length, frequency, con- of previously-uttered elements (Dell et al., 1997). By contrast, creteness/imageability, givenness, and other factors (Bock, 1982). young children, who are less experienced speakers, produce more Second, other action and motor planning processes show these errors overall and a relatively higher proportion of perseverations same Easy First tendencies. MacNeilage and Davis (2000) argued (Stemberger, 1989). The impact of these phenomena is three- that the distributional regularities of consonant and vowels in fold. First, they illuminate the demands of language planning, infants’ babbling and early words reflect infants’ tendencies to www.frontiersin.org April 2013 | Volume 4 | Article 226 | 3 MacDonald Language production shapes form and comprehension order segments more easily articulated within the infant vocal over time and over other intervening utterances. The effect is tract before more difficult ones. Similarly, research in navigation argued to be not (or not only) the temporary activation of recent shows an Easy First action ordering preference: when humans or plans but rather a manifestation of long-term implicit learning animals have to visit several locations along a path, they typically of syntactic structure (cf. Branigan et al., 1999; Chang et al., begin with the nearest one (Gibson et al., 2007); if that nearest 2006). On this view, language users are continually learning from one is made difficult to reach because of obstacles, then more their and others’ language use; with every utterance, a syntac- distant unobstructed locations tend to be visited first (Miyata tic plan becomes more likely to be used in the future. Thus, and Fujita, 2011). Similarly, when humans are describing routes while the phenomenon is often described as one of short-term through a network, they also tend to begin by describing the sim- repetition, its learning basis links it to retrieval from long term plest one first (Levelt, 1982). Third, Easy First biases in serial memory—whereas Easy First refers to the effect of retrieval of ordering inherently follow from computational models of action individual words on word order, Plan Reuse effectively refers to planning, in which alternative sub-plans compete for entrance the retrieval of the sentence structure itself. The two constraints into an action plan, via selection mechanisms in sequence plan- jointly exert their influence on utterance form: Even in lan- ning (e.g., competitive queuing, Grossberg, 1978; Hartley and guages with very free word order, allegiance to favored structures Houghton, 1996) or via gating functions of attention in models of (Plan Reuse) combines with Easy First in shaping utterance forms cognitive control, in which more practiced/easier elements, which (Christianson and Ferreira, 2005). require less attention, precede more difficult ones in a developing The reuse of at least partially lexically-independent abstract plan (Botvinick and Cohen, submitted). Thus, the Easy First bias plans is in some ways consistent with an autonomous syntac- in language production is not a stipulative principle or language- tic representation independent of semantics (Chomsky, 1957), specific phenomenon; instead it follows naturally from attested although the notion of adapting a prior syntactic plan to a new aspects of motor and action planning—that a plan precedes its utterance, and the notion of sentences, phrases, and words as execution, that planning is incremental, that the plan is hierar- plans and sub-plans, are less consistent with the contrast in gen- chical with subplans that must be ordered in some way, that plan erative linguistics between a stored lexicon vs. generative syntax. development entails retrieval from long term memory, and that Moreover, the reuse of abstract plans is not unique to language, this retrieval varies in speed and accuracy. as Plan Reuse appears in many non-syntactic and non-linguistic These results suggest that the way that utterance planning domains. Its effects are evident in recall from long term memory, unfolds over time has a substantial impact on the word orders in which people have a tendency to recall elements in the serial and sentence structures that language users produce. Moreover, order in which they have frequently occurred in past experience this work suggests a mechanistic basis for the observation that (Miller and Selfridge, 1950). There is also increasing evidence variation in language has functional importance (Givón, 1985): for structured non-linguistic stimuli such as action sequences Word order variation, such as active/passive forms (The noise affecting subsequent production of certain sentence structures, startled the boy vs. The boy was startled by the noise)and the suggesting that the re-use phenomena are not inherently linguis- English dative alternation (give Mary a book vs. give a book to tic (Allen et al., 2010; Kaiser, 2012). More broadly, similar Plan Mary) allows producers the freedom to place easily retrieved ele- Reuse appears in many non-linguistic motor behaviors in humans ments early, permitting early execution of the plan, and allowing and animals and is attributed to implicit motor learning. It is more time to plan the more demanding elements. Thus, in con- for these reasons that the reuse and adaptation of prior motor trast to Jackendoff’s (2002) suggestion that syntactic flexibilities plans for subsequent action is thought to be a hallmark of motor are vestiges of ancient protolanguage, before syntactic constraints planning and learning (Rosenbaum et al., 2007), and motor learn- became more rigid, the PDC holds that word order flexibility ing in the service of language appears to be no different. This has real value to language producers and emerges from action point reappears in the section Implications, Limitations, Future planning mechanisms that maximize fluency. Directions. Plan Reuse: a source of word order rigidity Reduce Interference Despite the enormous impact of Easy First on word order, it Whereas Easy First and Plan Reuse stem from ease of recall from cannot be the whole story—people’s utterances are not sim- long term memory, Reduce Interference reflects properties of ply strings of words ordered by ease of retrieval from memory. immediate memory instead of or in addition to long term recall. A Production also accommodates constraints on permissible word classic finding in verbal and non-verbal short-term memory tasks orders in a language. A second significant influence on utter- is that the to-be-remembered elements interfere with one another ance form also favors easy, more practiced plans, but in this in memory during the short interval between their presenta- case, what is easy is the abstract sentence plan itself rather than tion and recall, with increasing interference when the elements the word or phrase elements (sub-plans) within it. Producers share similarity in sound, meaning, spatial location, or other have a conspicuous tendency to reuse recently executed utterance dimensions (Conrad and Hull, 1964; Anderson, 1983). Because plans, so that the likelihood that a speaker utters a passive sen- utterance plans are maintained before execution, it is not surpris- tence, for example, increases if that speaker has recently heard, ing that elements in the plan also interfere with one another. For read, or uttered another passive sentence (Weiner and Labov, example, when two semantically related nouns must be planned 1983; Ferreira and Bock, 2006). This tendency toward Plan Reuse and uttered in close proximity (e.g., . . . the couch and the chair. . . ), (also called structural persistence or syntactic priming) persists utterances take longer to plan and contain more errors than when Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 4 MacDonald Language production shapes form and comprehension this similarity is not present (Smith and Wheeldon, 2004). These result from attempts to mitigate the computational demands of effects may be a consequence of winner-take-all production: The production planning rather than a specific discourse strategy to path from conceptual message to word selection includes the par- emphasize certain information for the comprehender, then we tial (unconscious) activation of many alternatives (couch, sofa, should also see effects of the third factor described above, Reduce loveseat, chair, furniture, etc.), and successful production requires Interference, interacting with the other two. This prediction is that only one of these enter the utterance plan. Having then set- also supported; Gennari et al. (2012) found that when the agent tled on couch and inhibited all others, the producer has additional and patient of an event are semantically similar (e.g. boy, girl), difficulty—interference—when it becomes necessary to retrieve people more frequently describe the patient in passive structures one of these inhibited options (e.g., chair; Tydgat et al., 2012). such as The girl was pushed (by the boy),in which theagent As with other examples discussed above, producers mitigate this (boy) is demoted to a by-phrase or eliminated entirely in agent- interference via choices of utterance form (Gennari et al., 2012). A less passives (The girl was pushed). Here the system mitigates the specific example is given in the next section, which considers how demands of production by omitting, delaying, or demoting sen- Reduce Interference interacts with Easy First and Plan Reuse. tence elements that are affected by memory interference. These results suggest that while producers may sometimes (consciously The three factors in action or unconsciously) select a syntactic structure such as passives Easy First and Plan Reuse can pull in opposite directions, as to convey a particular message, substantial variation in utter- Easy First promotes word order flexibility to allow easily-retrieved ance form stems from the degree to which certain choices can words before more difficult ones, whereas Plan Reuse promotes reduce production difficulty for the producer. The results also rigidity of word order via re-using previously uttered struc- suggest that both word order variation and word order rigidity tures. Cross-linguistically, many distributional patterns of word have real value in production planning. Both the tendency to lead orders reflect this tension, owing to different degrees of word with easy elements and the tendency to adopt well-worn sentence order flexibility in different languages. In English and many types emerge from the nature of learning and retrieval from long other languages, passive structures such as (1b) are more com- term memory, in that highly frequent elements or well-practiced mon with animate subjects (boy) than with inanimate subjects abstract plans, are preferred over more attention-demanding like window. This pattern follows straightforwardly from the alternatives. On this view, implicit choices of both lexical items greater ease of memory retrieval of animate nouns than inan- and sentence forms are shaped by the same memory-retrieval imate ones (Bock, 1982), and from the fact that the passive constraints. construction allows the easily-retrieved noun boy to be placed Beyond increasing the fluency of an individual’s utterances, early in the sentence and in a prominent position (sentence these three production biases have another important conse- subject). quence at the heart of the PDC that “individual-level behaviours result in population-level linguistic phenomena” (Scott-Phillips 1a. Active: The ball hit the boy hard, but he was OK. and Kirby, 2010, p. 1364). Summed over millions of utterances 1b. Passive: The boy was hit hard by the ball, but he was OK. and many language producers, implicit production choices favor- ing less-difficult forms create dramatic statistical regularities in The three production planning factors make testable predictions language usage linking conceptual messages, words, and sen- about variation in passive use. First, if the relationship between tence types. The next section relates this perspective to other noun animacy and active/passive form is the result of utterance approaches to language typology and universals and argues that a planning pulled between Plan Reuse (favoring the more com- greater attention to production processes offers insight into ques- mon Active form) and Easy First (favoring early mention of tions about why some distributional patterns are more frequent animates), then we would expect that animacy/Easy First effects than others. on structure would be smaller in those languages (such as Slavic languages) that have a strong bias to use active forms. Results DISTRIBUTIONAL REGULARITIES AND LANGUAGE of this type are perhaps not surprising, because by definition, TYPOLOGY a strong allegiance to a single dominant word order to convey Functional linguists, language typologists, and historical linguists a particular message will allow less room for word order flexi- investigate the distributional regularities across the world’s lan- bility to accommodate ease of retrieval (Myachykov et al., 2011; guages and their change over time, with one goal being the Gennari et al., 2012) . Second, utterance planning time should identification of significant cross-linguistic tendencies or lin- increase when these forces conflict compared to situations when guistic universals that could illuminate the nature of human they converge on the same form. This prediction is also supported 3 language . Many functional linguists point to language use as (Ferreira, 1994). Third, if these structure and word order choices a source of cross-linguistic patterns, meaning that languages tend to have (or develop over time) properties that serve the See Bresnan and Ford (2010), Stallings et al. (1998),and Wasow (1997),for needs of language users (see Bybee, 2006,for review). The other examples of the tension between Easy First and Plan Reuse, though not using these terms. The reasons why one language would have freer word order than another is of course something to be explained within any perspective. In This statement dramatically simplifies functional, historical, and typological production based approaches, large-scale corpus studies should prove useful, linguistics as well as the debate about whether there are truly universals of as in investigations of the rigidity of use of dative constructions in American language or merely asymmetries in the distribution of language features in vs. Australian English (Bresnan et al., 2007; Bresnan and Ford, 2010). the world’s languages (Evans and Levinson, 2009). www.frontiersin.org April 2013 | Volume 4 | Article 226 | 5 MacDonald Language production shapes form and comprehension PDC’s view is related, but it holds that specifically the needs of language production, memory recall, action planning, and motor producers have the most direct effect on patterns of sentence control. structure. Indeed, this article is by no means the first to sug- gest that utterance planning processes in language production PRODUCTION GETS TYPOLOGY WRONG have an important role shaping language typology and histori- In English and many other languages, shorter words and phrases cal change (e.g., Bock, 1982; Bock and Warren, 1985; Jäger and tend to precede longer ones, which has been attributed to Easy Rosenbach, 2008). Typologists have long observed that linguis- First (Stallings et al., 1998). However, Hawkins (2004) describes tic variation happens for a reason (e.g., Givón, 1985), and the some notable exceptions in Japanese and other “head-final” lan- production processes described above take steps toward a more guages, casting doubt on production as a source of typological mechanistic account of why word orders may vary from one sit- patterns of word order: “The preference for long before short uation to the next as a function the nature of retrieval from in Japanese is not predicted by current models of language pro- long term memory, the role of attention in gating competing duction, all of which are heavily influenced by English-type processes in utterance planning, memory interference among [languages]” (p. 110). Hawkins is correct that language produc- entities in the utterance plan, and other factors. Moreover, the tion researchers have investigated relatively few languages, and learning mechanisms of production that promote reuse of prior this concern is compounded by psycholinguists’ tendency to pur- plans are an obvious candidate for informing accounts of lan- sue narrow, controlled studies focusing individually on Easy First, guage change (e.g., Bybee and McClelland, 2005). Despite this Plan Reuse, or Reduce Interference, with relatively little atten- potential synergy between a more detailed study of language tion to the fact that multiple factors can contribute to retrieval production mechanisms and language typology, there is rel- from memory and thus word order. Things are improving on atively little consideration specifically of language production both these fronts, and more recent production accounts do inves- processes in the typology and universals literature (though see tigate the origin of opposing effects of phrase length in English Bybee, 2006;also Jäger and Rosenbach, 2008, and associated and Japanese (Yamashita and Chang, 2001), including a computa- commentaries). Why not? One obvious answer—lack of inter- tional model of language production that develops Short-before- action between language typologists and language production Long preferences when trained with English input and learns a researchers—is generally true but not fully satisfying, because it Long-before-Short preferences when exposed to Japanese input doesn’t address why researchers in these areas feel little motivation (Chang, 2009). Analyses of the model’s performance in the two to interact. Five more substantive assumptions underlying the dis- language environments point to Plan Reuse as one important connection are considered here, together with arguments for a force in developing the ordering preferences, in that tendencies rapprochement. for ordering object and recipient noun phrases reflect the adapta- tion of plans from more common sentences with only one noun phrase. Thus, production work may have been late to the party CONCEPTUAL REPRESENTATIONS EXPLAIN LANGUAGE FORM here, but if Chang is correct about the role of learning mecha- WITHOUT LANGUAGE PRODUCTION MECHANISMS nisms and Plan Reuse, then his mechanistic account of utterance Linguists have long noted that more conceptually salient ele- planning will prove central to these cross-linguistic differences. ments (i.e., those more important to the producer and/or comprehender) occur early in utterances, such as the ten- dency for agents/animate entities to appear before undergoers of THE GRAMMAR-PERFORMANCE DISCONNECT action/inanimate entities. These and related patterns have been The value of distinguishing linguistic competence (knowledge: attributed to varied forces, such as an Agent First principle in the grammar) and performance (use) has long been a source of Universal Grammar (Jackendoff, 2002) and functional accounts debate within linguistics (Newmeyer, 1998; Jackendoff, 2007)and in which elements that are salient in the discourse receive a promi- is beyond the scope of this article. Two trends are worth not- nent sentence position (Chafe, 1976; Goldberg, 2006). This latter ing, however: First, some linguistic approaches increasingly view position clearly shares a good deal with Easy First, but the pro- grammar itself as a graded representation emergent from experi- duction account goes beyond an appeal to salience in important ence with language tokens (Bybee, 2006; Bresnan et al., 2007), and ways. First, the Easy First bias in production grounds the con- this position (whether or not “grammar” is invoked in the expla- ceptual salience effect in ease of recall from long term memory. nation), is central to recent production-based accounts of word Since salience itself is not acting directly on prominence but order variation (Kuperman and Bresnan, 2012). Second, whatever rather via ease of recall, this approach correctly predicts that other one’s position on the nature of grammar, utterance planning itself non-salience factors affecting ease of recall (e.g., word length) clearly shapes utterance form, and as such, production merits can also affect word order. Second, the incremental nature of more attention in typology if only to better attribute distributions motor planning for production explains why the privileged loca- to the work of grammar vs. performance. tion for easily recalled entities is early in the utterance plan Relatedly, the culture of controlled laboratory studies in lan- rather than saving the easiest for last. And third, filtering con- guage production is at odds with typologists’ interests in the ceptual salience through the production system accounts for broad sweep of cross-linguistic patterns. These trends are chang- situations in which communicative goals influence salience (via ing in several ways. First, researchers are increasingly investigating task-specific allocation of attention) rather than the other way the link between individual-level phenomena (as studied in many around (Kuchinsky et al., 2011). Thus, “salience” does affect language production studies) and the population-level phenom- word order, but it gains external validity via an understanding of ena, where interactions among many individuals affect language Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 6 MacDonald Language production shapes form and comprehension form over time (e.g., Scott-Phillips and Kirby, 2010). Clearly the to make input more regular, as when a child uses holded for the PDC will benefit from improved understanding of individual- past tense of hold rather than held. Considered from the point population interactions. Second, there is increasing use of large of production, the use of holded vs. held is an implicit choice corpora as a form of production data being brought to bear on of utterance form over available options, similar to saying cat typological issues (e.g., Piantadosi et al., 2012), and a move by vs. kitty or a passive vs. active sentence. However, a child who some psycholinguists to adopt information theoretic approaches utters holded may have never encountered it before; what could to language performance, in which accounts of language distri- cause a bias toward producing a form that’s generally unattested butions invoke notions of communicative efficiency (e.g., Jaeger in the input? Ease of production is a good bet, both because and Tily, 2011). This perspective obviously has clear overlap with it is such a powerful force in adult language production and functional linguistic accounts of typology, but from the PDC per- also because child utterances are full of omissions and other spective, it’s important to incorporate a detailed account of the simplifications that reduce utterance difficulty despite being unat- cognitive control and memory demands of production planning tested in the adult input. The production force that promotes into accounts of “efficiency” and why some distributional pat- overregularization is Plan Reuse, where the abstract plan here is terns are more common than others. It remains to be seen how the regular inflection, which becomes increasingly common as this tension between information theoretic and more mechanistic the child learns more verbs (Rumelhart and McClelland, 1986). accounts plays out. On this view, children’s novel forms such as omitting an initial unstressed syllable (e.g., saying nana for banana)and holded for TYPOLOGY AND UNIVERSALS REFLECT ACQUISITION, NOT held are not distinct phenomena but are instead both examples PRODUCTION of the strong influence of Plan Reuse, reflecting the dominance Whereas many functional linguists consider adult language use of first syllable stress in English nouns (Echols and Newport, as key to understanding language universals and change (Bybee, 1992) and the dominance of the regular inflectional paradigm in 2006), others point to the child learner as the engine of change, English. either via the application of Universal Grammar (e.g., Lightfoot, This example argues for a different approach to considering 1999) or via innate learning biases in the child (Hudson Kam the role of learning in language typology and change, namely and Newport, 2009; Culbertson et al., 2012). From the PDC per- shifting the question from “Why are some forms more easily spective, this distinction between acquisition and use is a false learned by the child?” to a child version of the same question one, because learning in production (and comprehension) never we’ve asked about adults: “Why are some forms more often stops; there is not first a phase of acquisition and later one of produced?” Viewing the overregularization effect as owing to pro- use (Seidenberg, 1997; Chang et al., 2006) . Learning specifically duction choices is broadly consistent with accounts in which the in the service of production may be particularly important in effects of experience with individual words and with the regu- understanding the nature of language typology and change, first lar paradigm (the plan) vary with the amount of prior exposure because utterance planning requires memory retrieval, which is (Rumelhart and McClelland, 1986). This approach also yields the itself a powerful determinant of further learning (Karpicke, 2012). correct prediction that children who overregularize may nonethe- Thus, each production event should have on average a stronger less comprehend the irregular form that they don’t produce effect on subsequent learning than each comprehension event, (Clahsen et al., 2007; Ramscar and Yarlett, 2007). These results owing to the greater retrieval demands in production. Second, suggest there may be real value in considering the child’s utter- production is winner-take-all, meaning that someone who has ance planning demands in phenomena that have previously been perceived variable input in the past must commit to only one attributed to more general learning biases or Universal Grammar. form for any given utterance, with potential consequences both This position is clearly not anti-learning but rather an argument for subsequent distributional patterns and for learning over one’s for considering what the learning is for. own productions . Thus, more attention to learning for produc- tion could inform current unknowns concerning hypothesized LANGUAGE IS TAILORED TO THE COMPREHENDER links between acquisition and typology: “why learners acquire In arguing for a central role for production in shaping language certain types of patterns more easily than others (and why lan- form, the PDC does not deny that other forces may also influ- guages therefore more commonly exhibit these patterns)” (Aslin ence form. However, if these other forces turn out to be extremely and Newport, 2012, p. 174). powerful, they could erode the PDC’s claim for the centrality A central assumption in the role of learning biases in lan- of production in shaping language form. One alternative force guage acquisition and typology is that child learners are biased shaping utterance form is audience design,the idea that language producers tailor their utterances to accommodate the needs of An approach favoring continuity between acquisition and use raises ques- the comprehender. Clearly producers do make adaptations to the tions about discontinuities, specifically sensitive period effects in language needs of the perceiver; the act of language production is itself an acquisition. Proactive interference approaches, in which prior learning affects accommodation, in that the producer is adapting to the fact that the rate of subsequent learning (Seidenberg and Zevin, 2006), suggest that the perceiver is not a mind reader and needs an overt linguis- child vs. adult language users may have different effects on language form, but tic signal. However, this adaptation is inherently limited, because not because of a distinction between acquisition and use. 5 the producer is also not a mind reader and therefore cannot fully Frequency boosting, in which a slightly dominant form strengthens over time know what needs the perceiver has. Thus, both audience design (Singleton and Newport, 2004), should be a natural consequence of winner- take-all production and learning from one’s own utterances (Plan Reuse). and production-driven utterance choices likely exist in parallel. www.frontiersin.org April 2013 | Volume 4 | Article 226 | 7 MacDonald Language production shapes form and comprehension Attributing aspects of utterance form to producer needs (as is not whether there is audience design (there is), but rather in the PDC) vs. audience design is a complex undertaking, for how distributional patterns emerge from the specific computa- several reasons. First, the most obvious potential evidence for tional demands of language production as executed by producers audience design—that comprehenders benefit when producers who have a communicative goal. Because audience design con- use certain forms and find other forms difficult—turns out not to tributes to the computational demands of utterance planning, be that useful. Perceivers routinely benefit from statistical patterns researchers who study the mechanisms of production planning in their input that have no audience design; we can predict the tra- should accommodate theaudiencedesignliterature morefully, jectory of a bouncing ball, but the ball does not aim to help our and vice versa. tracking. Moreover, producer variation that might seem designed SUMMARY for an audience can instead have purely production-centered explanations. Consider the phenomenon of phonological reduc- This highly selective discussion contains almost no typological data and omits many issues in the current literature. The aim is tion, in which speakers introduce a word into a discourse with a fairly careful articulation and later re-mention it in a less precise not to review PDC contributions to typology (that section would be very short indeed) but rather to suggest that there is sufficient “reduced” form (Fidelholtz, 1975). Listeners clearly have learned these reduction patterns and benefit from them (Dahan et al., promise for cross-disciplinary interaction, specifically that the computational demands of utterance planning, and producers’ 2002), but this benefit does not mean that phonological reduction is designed to help the listener. Indeed, reductions of this sort are attempts to minimize them, should be investigated further as an important driving force in cross-linguistic language typology and an inherent consequence of motor learning and practice gener- change. Though not elaborated here, the reverse is also true: Work ally, including in motor behaviors with no audience (Müller and Sternad, 2004). on the statistical distributions in the world’s languages, and the way that languages change over time, can inform psycholinguistic A second complicating factor is that audience design is not cost-free: accommodating a comprehender will itself impose accounts of language processes (e.g., Feist, 2010; Culbertson et al., 2012). demands on the producer. The producer must work to iden- tify perceiver needs, and as this task becomes more difficult (e.g., requiring more elaborate inferencing), the amount of audi- COMPREHENSION CONSEQUENCES IN THE PDC ence design in the utterance declines (Horton and Keysar, 1996; Having reviewed implicit choices of utterance forms and con- Bard et al., 2007). Thus, we can’t talk about whether utterance sequences for distributions in the language input, we now con- form comes from perceiver or producer needs, because perceiver sider what comprehenders do with these distributions. The next accommodation inherently creates demands on the producer. section addresses what language users learn about distributional Third is the existence of accommodation in the other direc- regularities, and the two sections after that review two classic tion, in that perceivers accommodate the needs of the speaker examples of sentence comprehension phenomena, for which pop- (Duran et al., 2011). Some obvious examples, many of which are ular theories have attributed comprehension behavior to archi- not always identified as producer accommodation, include vari- tectural properties of the comprehension system—in effect, that ous forms of phonetic adaptations (Kraljic et al., 2008)and ambi- comprehension works the way it does because innate parsing guity resolution at lexical, syntactic, and other levels (MacDonald biases make it so. By contrast, the PDC approach suggests that et al., 1994). Perceivers may often be good at accommodation the comprehension results reflect distributional regularities in the because they have direct information about the producer’s per- language, which themselves can be traced to the joint actions of spective from the utterance itself. Indeed, recent information- Easy First, Plan Reuse, and Reduce Interference shaping the forms theoretic analyses have suggested that overall communicative of utterances during utterance planning. efficiency is not optimal when the producer is maximally clear and redundant, which would make utterances longer and more DISTRIBUTIONAL REGULARITIES AND PREDICTION IN carefully articulated than the perceiver needs. Instead, commu- COMPREHENSION nicative efficiency is higher when the producer uses short ambigu- Linguistic signals unfold over time, creating long distance depen- ous words and permits phonological reduction and substantial dencies, where the interpretation of some input is dependent additional ambiguity (Piantadosi et al., 2012). This arrangement on previous or upcoming parts of the signal. Integrating over works because comprehenders are so good at ambiguity resolu- these dependencies involves use of probabilistic information in tion and other forms of speaker accommodation. Such results both forward and backward directions to settle on the most likely turn the notion of audience design on its head: Tuning the conver- interpretation of the input. In the backwards direction, recently- sational interaction primarily to the producer’s needs, and letting encountered information allows further refinement of the earlier the perceiver accommodate the producer, is in a broad sense a input (MacDonald, 1994), with some effects strong enough to form of audience design: The producer adopts utterance forms affect what perceivers report they hear (Warren and Sherman, mitigating difficulty and maximizing fluency so that the conversa- 1974; Connine and Clifton, 1987; Mack et al., 2012). Use of statis- tion proceeds efficiently, without bogging down the process with tical dependencies in the forward direction does not necessarily more redundancy than the perceiver needs. entail exactly predicting upcoming words but rather generating These considerations suggest that audience design is not general expectations about grammatical category, gender, and incompatible with producers’ implicit choice of utterance forms other properties that greatly narrow the scope of possibilities that mitigate production difficulty. On this view, the key question (Van Petten and Luka, 2012). The notion that comprehenders are Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 8 MacDonald Language production shapes form and comprehension generating expectations for upcoming input has been a contro- This approach accorded well with the fact that, with few excep- versial one, as ithas notalways been clear thatpredictions could tions (Fodor and Inoue, 1994; Altmann et al., 1998), the lexi- be sufficiently constraining or efficiently computed (MacDonald cal content of sentences like (2) has minimal effect on English and Seidenberg, 2006). However, partial predictions can emerge speakers’ strong bias in favor of local modification, making verb naturally from a system operating under time pressure (Allen modification ambiguities the best available evidence for lexically- and Seidenberg, 1999), and predictions may arise from many independent innate parsing algorithms. correlated language statistics, so that low levels of the linguis- As Table 1 summarizes, the PDC approach accounts for the tic signal, such as acoustic or orthographic form, can provide local interpretation biases without innate parsing algorithms. extremely early probabilistic information about grammatical cat- Instead the effects stem from learning over the distributional reg- egory or syntaxeveninadvance of word recognition (Dikker ularities in the language, which in turn stem from the biases of et al., 2010). Moreover, parts of the signal that have predictive producers to favor certain sentence forms that minimize produc- value for upcoming percepts not only speed the processing of tion difficulty. the predicted elements downstream but may themselves be pro- In Step 1 in the table, the Easy First production bias dis- cessed more rapidly than uninformative signal, owing to cortical courages production of distant modification sentences like (2c) feedback mechanisms gating attention toward potentially infor- because more easily planned alternatives exist. In (2c), a relatively mative input (O’Brien and Raymond, 2012). Together, this work long phrase (that his cousins left)precedes a shortone (yester- reflects a point that’s evident in information theoretic accounts day), but Easy First promotes a short-before-long phrase order, of language processing but hasn’t consistently penetrated other as in John said yesterday that his cousins left,or Yesterday, John comprehension approaches, that there is always ambiguity in the said that his cousins left. Step 2 identifies the distributional con- language signal as it unfolds over time, and uncertainties about sequences of speakers avoiding utterances like (2c): Ambiguous both the upcoming and the recently encountered signal are a sentences like (2a) typically have a local modification interpre- source of processing difficulty (e.g., Hale, 2006; Levy, 2008). Use tation like (2b). Comprehenders are extremely sensitive to these of distributional regularities to reduce this uncertainty is a pow- statistics (Step 3), and they have difficulty comprehending largely erful advantage in comprehension, and the next two examples unattested forms like (2c), but they readily comprehend a spe- suggest how a deeper appreciation of this fact, together with an cial type of distant modification sentences that don’t violate Easy understanding of how production processes create certain dis- First and that do exist in the language. These results suggest that tributional regularities, reframes our understanding of sentence rather than an innate comprehension bias for local modification, comprehension. perceivers have a learned bias toward what has happened in the past, and that this prior linguistic experience owes to aspects of production planning. REINTERPRETING PARSING PRINCIPLES: VERB MODIFICATION This claim for the role of past experience on subsequent com- AMBIGUITIES prehension processes is at the heart of constraint-based accounts A pernicious type of sentence ambiguity, the verb modification of language comprehension, which have been applied to many ambiguity, is shown in (2), in which an adverbial phrase could modify one of two different actions described in the sentence. Example (2a) shows a fully ambiguous structure, (2b) shows an Table 1 | Production-Distribution-Comprehension (PDC) account of example in which verb tense disambiguates the sentence in favor greater comprehension difficulty for ambiguities resolved with of a the local modification interpretation in which the adverb yes- distant modification (2c) than with local modification (2b). terday modifies the nearest verb left rather than the more distant phrase will say, and (2c) is an example of distant modification, in PDC STEPS which tomorrow modifies the distant verb, will say. 1. Production: Easy First, where shorter phrases precede longer ones, discourages production of ambiguous structures like (2a) with intended distant modification, and instead promotes production of 2a. Verb Modification Ambiguity: John said that his cousins left other forms to convey the same message (MacDonald, 1999; yesterday. MacDonald and Thornton, 2009). 2b. Local Modification: John will say that his cousins left yester- 2. Distribution: As a result, ambiguous sentences with intended distant day. modification are much rarer than ambiguous sentences resolved with 2c. Distant Modification: John will say that his cousins left local modification (Sturt et al., 2003; MacDonald and Thornton, 2009). tomorrow. 3. Comprehension: The comprehension patterns reflect the language statistics in Step 2: (a) Overall, the rarer distant modifications are harder than the more English comprehenders greatly favor local modification (2b) over common local modification sentences (Altmann et al., 1998; distant modification (2c). This pattern is often thought to arise MacDonald and Thornton, 2009). directly from innate parsing or memory biases to favor local (b) However, a subtype of verb modification ambiguities don’t violate phrasal relationships over long distance ones, variously formu- Easy First in their distant modification form, owing to the relative lated as Right Association (Kimball, 1973), Late Closure (Frazier, length of phrases in these sentences. These are readily produced 1987), and Recency (Gibson et al., 1996). A key assumption by speakers who intend distant modification, are common in the has been that these parsing principles operate on purely syntac- language, and are easily comprehended (MacDonald and Thornton, tic representations without lexical content (e.g., Frazier, 1987). 2009). www.frontiersin.org April 2013 | Volume 4 | Article 226 | 9 MacDonald Language production shapes form and comprehension other syntactic ambiguities (see MacDonald and Seidenberg, 3b. Subject relative: The woman [who yelled at me] said I’d have 2006, for review). The added value of the PDC is (a) a greater to pay for the broken glass. emphasis on the role of learning probabilistic constraints, and (b) an account of the production basis of the language distributions The origin of relative clauses’ importance can be traced to claims that people learn and use to guide comprehension. Extending by Miller and Chomsky (1963) concerning reasons behind the the PDC to other syntactic ambiguities is ongoing; the approach comparative difficulty of subject relatives vs. center embedded holds promise because (a) these ambiguities turn on the rel- object relatives. Chomsky and Miller (1963) observed that the ative frequency of alternative uses of language, which can be repeated recursive operation of embedding one object relative readily learned from input (Wells et al., 2009), and (b) certain inside another one yields an uninterpretable sentence; their exam- production choices are known to affect syntactic ambiguity. For ple was The rat [the cat [the dog chased] killed] ate the malt. Miller example, variation in availability of genitive forms (the profes- and Chomsky (1963) viewed the difficulty of these sentences sor’s daughter vs. the daughter of the professor)in English vs. as following from a distinction between linguistic competence other European languages affects the distribution of noun mod- and ability to use that knowledge—linguistic performance. They ification ambiguities and their interpretation in these languages argued that while linguistic competence (here, recursion) is infi- (see Mitchell and Brysbaert, 1998,for review and Thornton nite, performance, specifically the ability to use this knowledge to et al., 1999 for constraint-based studies of cross-linguistic simi- comprehend center embedded structures, is constrained by lim- larities and differences). Similarly, producers manage production itations on short-term memory capacity (Miller, 1956). In the demands through the use of optional words (e.g., Ferreira and case of object relative clauses, the memory burden stems from Dell, 2000), which have substantial effects on ambiguity, the dis- the multiple incomplete noun-verb dependencies arising as the tribution of form-meaning pairings, and consequent experience- sentence unfolds, so that the comprehender must first anticipate driven ambiguity resolution processes. Thus, the PDC prediction a verb for each noun (the rat the cat the dog)and hold these is that all syntactic ambiguities can ultimately be traced back unintegrated nouns in memory, and then when the verbs are to producers’ implicit utterance choices (many in the service of encountered (chased killed ate), associate them appropriately with reducing utterance planning difficulty), the consequent distribu- the nouns (Wanner and Maratsos, 1978; Gibson, 1998). By con- tions in the language, and comprehenders’ learning over those trast, the more comprehensible English subject relatives interleave distributions. nouns and verbs, reducing the memory burdens: The dog [that chased the cat [that killed the rat]] howled. REINTERPRETING SYNTACTIC PARSING AND WORKING MEMORY It is difficult to overstate both the impact of Miller and BURDENS Chomsky’s analysis and the subsequent reach of relative clauses The next example, relative clause interpretation, repeats the PDC into virtually all corners of language comprehension research. argument—mitigating production difficulty leads to utterance Several additional factors have contributed to the central posi- choices that lead to distributional regularities that lead to com- tion of relative clauses in theories of memory and language use. prehension patterns. Relative clauses nonetheless merit detailed First, relative clauses are widely held to be syntactically unam- attention, first because they illustrate complex interactions of all biguous (Babyonyshev and Gibson, 1999), so that comprehension three production biases, and second because they have played an difficulty can’t be attributed to ambiguity resolution processes. outsized role in theories of both syntax and language compre- Second, subject and object relatives can be made to differ by hension, so that a revision of traditional claims has substantial only the order of two phrases, as in the order of the senator consequences. and attacked in (4a,b), so that researchers can contrast com- prehension of sentences for which the lexical content seems perfectly matched. The vast majority of a very large number The relative clause trifecta: recursion, competence-performance, and working memory of studies in English and many other languages, across chil- dren, adults, individuals with brain injury, disease, or devel- Relative clauses are noun modifiers that include a verb, as in opmental atypicality, show that object relatives are more dif- examples (3a,b). In (3a) the ball is being modified by the brack- ficult than their matched subject relatives (see O’Grady, 2011, eted relative clause; because the ball is the object of the relative for review). The logic here seems perfectly clear: Because the clause verb (threw), this structure is called an object relative (or difference in difficulty can’t be ascribed to lexical factors or center-embedded) clause. A subject relative clause is illustrated in ambiguity resolution, it must reflect purely syntactic opera- (3b), where woman is the subject of the relative clause verb yelled. tions and the memory capacity required to complete them These two examples seem pretty innocuous, but in fact subject (Grodner and Gibson, 2005). and object relative clauses have played a central role in defining the differences between language competence and performance in generative linguistics, and they also have had an enormous 4a. Object relative: The reporter [that the senator attacked] impact in essentially every area of comprehension research, from admitted the error. acquisition, to adult comprehension, to studies of aphasia and 4b. Subject relative: The reporter [that attacked the senator] other language impairments. admitted the error. 3a. Object relative: The ball [that I threw to Harold] went over This competence-performance account of working memory his head and broke a window. overflow in relative clause comprehension continues as the Frontiers in Psychology | Language Sciences April 2013| Volume4| Article226 | 10 MacDonald Language production shapes form and comprehension Table 2 | PDC account of greater comprehension difficulty for object dominant perspective in linguistics, language acquisition, adult than subject relative clauses (citations refer to English results). psycholinguistics, and communicative disorders, despite crit- icisms of each of the components of this argument. These 1. Object relatives (5a) are common when the noun being described is criticisms include evidence that multiply center-embedded sen- inanimate (toy) but are avoided when the relative clause describes tences need not be incomprehensible (Hudson, 1996), com- something animate (boy), passive relatives (5b) are produced instead prehension difficulty is strongly influenced by the words in (Montag and MacDonald, 2009; Gennari et al., 2012). These patterns the sentence and therefore cannot reflect purely syntactic pro- owe to at least three production biases: cesses (Traxler et al., 2002; Reali and Christiansen, 2007), (a) Easy First: animate nouns are conceptually prominent and easily object relatives do contain a non-trivial amount of ambigu- retrieved from memory, leading to their position in early or ity directly related to comprehension difficulty, again refut- prominent sentence positions. The passive relative (5b) allows the described noun to be in the prominent subject position of the ing the assumption that relative clauses provide a pure mea- relative clause. sure of syntactic difficulty (Gennari and MacDonald, 2008), (b) Plan Reuse: the rate of passive relatives varies with the viability of the degree of prior experience with object relatives predicts passives in the language more generally, reflecting the reuse of comprehension success in children and adults, a result not passive forms from other sentence types (Montag and captured by memory overload approaches (Roth, 1984; Wells MacDonald, 2009). et al., 2009), people’s comprehension capacity for recursive struc- (c) Reduce Interference: there is more interference between tures is more accurately described by a system in which work- conceptually similar entities [e.g. boy/girl in (5)] than when an ing memory is inseparable from linguistic knowledge than by animate entity (girl) acts on an inanimate one (toy). This one with separate competence and performance (Christiansen interference can be reduced by omitting the agent in the utterance and Chater, 2001), and that cross-linguistically, relative clause plan, which is possible in passive forms (5b), but not in object complexity does not always predict comprehension difficulty relatives (5a). The higher the conceptual similarity between (Lin, 2008; Carreiras et al., 2010). Theresilienceof mem- sentence participants in the event to be described, the more speakers produce passive agent-omission relative clauses ory overflow accounts in the face of these myriad challenges (Gennari et al., 2012). in part reflects the essential usefulness of the constructs of 2. People readily learn these correlations between animacy and relative working memory capacity and competence-performance dis- clause type (Wells et al., 2009). tinctions in cognitive science. However, a second factor is that 3. Comprehenders who encounter the start of a relative clause have there has been no really compelling alternative account that very different expectations for how it will end, depending on whether captures both the subject-object relative asymmetry as well as something animate or inanimate is being described, with these other phenomena. The PDC approach aims to provide consequences for comprehension: exactly this. (a) When relative clauses describe something inanimate like toy, English speakers rapidly anticipate an object relative (5a); for The relative clauses that people produce animates (boy), object relatives are vanishingly rare and are not Insight into why object relatives are hard requires noting pro- expected by comprehenders (Gennari and MacDonald, 2008). ducers’ available choices, specifically that there are two ways to (b) The less producers are willing to say an object relative to convey a describe the patient/theme of some action with a relative clause, particular message, the less comprehenders expect one, and the either an object relative (5a) or a passive relative (5b; curly brack- more difficult the comprehension is when a sentence in fact turns out to contain an object relative clause (Gennari and MacDonald, ets indicate the optional passive “by-phrase” identifying the agent 2009). of the action). 5a. Object relative: The boy/toy [that the girl splashed] was are produced less often when describing animate entities than dripping wet. 6 inanimate ones . 5b. Passive relative: The boy/toy [that was splashed {by the girl}] Figure 1 also shows large cross-linguistic differences in the was dripping wet. overall rate of object relative use. The reasons behind these differ- ences are quite complex and of course reflect important topics in Step 1 in Table 2 describes how producers’ use of object rel- language typology. Some variation in overall tolerance for object atives vs. passive relatives is shaped by the joint action of relatives appear to reflect Plan Reuse and the viability of passives Easy First, Plan Reuse, and Reduce Interference. When English in main clauses in a language (Montag and MacDonald, 2009), producers are describing something inanimate (e.g., toy), they and other important factors may include whether the language readily produce object relatives like (5a), but they almost has other utterance forms that speakers might use beyond object never do this to describe something animate (boy). Instead, they utter passive relatives like (5b). This pattern is not lim- In all Figure 1 studies, native speakers answered questions about cartoon pic- ited to English; my colleagues and I have investigated rela- tures in which animate agents acted on animate or inanimate entities (e.g., a tive clause production in six languages, which differ widely girl in a pool splashing a boy or a toy). Critical questions required speakers in word order in main and relative clauses, the amount of to describe the objects of actions; e.g., What is green? referred to the toy being case marking on nouns, the availability of alternative struc- splashed. Participants were not explicitly instructed to use relative clauses but tures to express this same message, and many other proper- frequently did so, with replies such as The toy that the girl is splashing/that’s ties. Figure 1 shows that in all six languages, object relatives being splashed {by the girl}. www.frontiersin.org April 2013 | Volume 4 | Article 226 | 11 MacDonald Language production shapes form and comprehension On this view, relative clauses, which have been central to cur- rent conceptions of memory and language use in virtually every subfield of psycholinguistics, turn out to be wholly unsuited for that role, as they are not unambiguous, and their comprehen- sion reflects detailed knowledge of correlations between words and structures, not abstract syntactic representations. What then becomes of working memory limitations as a source of com- prehension difficulty, particularly within Miller and Chomsky’s (1963) competence-performance claims for infinite recursion limited by working memory? The short answer is that researchers may further debate competence-performance distinctions, but relative clauses should no longer be offered as evidence of over- flow of syntactic memory representations that limit infinite recur- sive capacity. A more precise answer about implications of the FIGURE 1 | The frequency with which object relative clauses are relative clause work requires closer attention to what working produced to describe animate and inanimate entities in a picture memory is and isn’t. In saying that the PDC account refutes description task, calculated as a percentage of all relative clauses produced. The English, Spanish, and Serbian data are from Experiments claims for working memory limitations in sentence comprehen- 1a,2,and3of Gennari et al. (2012), respectively. The Japanese data are sion, my colleagues and I do not mean that working memory from Montag and MacDonald (2009), Korean from Montag et al. doesn’t exist—to the contrary, a prime reason why language users (in preparation), and Mandarin from Hsiao and MacDonald (in preparation). track the statistics of the language and use them to anticipate upcoming input is precisely because language comprehension and passive relatives, and the extent to which nouns are marked requires significant memory capacity, and generating expecta- for case and the flexibility of word order in a language, both tions for likely outcomes reduces these burdens. However, we of which appear to modulate the degree interference between do reject the notion that people’s working memory capacity can the agent of the relative clause and the entity being described be described as a performance limitation independent of their by an object relative clause (Gennari et al., 2012). There are linguistic knowledge/competence (MacDonald and Christiansen, undoubtedly other complex influences as well. 2002; Acheson and MacDonald, 2009; Wells et al., 2009). Our Although we are just beginning to understand the factors position reflects broader trends linking working memory and behind the patterns in Figure 1, it is clear that speakers’ very dif- long-term knowledge (Cowan, 2005), emergent from the tem- ferent choices for animate-describing and inanimate-describing porary maintenance needs of other cognitive processes (Postle, relative clauses have robust effects on the distributional regular- 2006). Specifically for relative clauses, comprehension capacity ities in these languages. Steps 2–4 of Table 2 show the cascade varies with long term knowledge of these structures, derived from of consequences of these choices—comprehenders rapidly learn experience. Language producers provide some kinds of experi- the robust form-meaning correlations (Step 2), and they bring ences (some kinds of relative clauses) more than others, with this information to bear in comprehension, such that they consequences for language distributions, learning over those dis- expect object relatives where they’re commonly produced but tributions, and for the memory demands needed to comprehend are surprised by them in unexpected environments, leading to these structures—the memory capacity and experience cannot be comprehension difficulty (Step 3). The vast majority of studies separated. Of course computational limitations, including mem- demonstrating object relative processing difficulty have used ory limitations, are also at the heart of the PDC argument for materials in which something animate is being described—the why producers prefer some utterance forms over others, but this very situation that producers avoid and that comprehenders does not mean that the competence-performance distinction can have learned not to expect. Together, the steps in Table 2 suggest simply be shifted to production, because again, linguistic work- that object relative clause comprehension is simply another ing memory, specifically the capacity to produce certain utterance example of ambiguity resolution—comprehenders are “led down forms, is not separate from long term linguistic knowledge or the garden path,” as the saying goes in parsing research, by experience (Acheson and MacDonald, 2009). relying on past experience that leads to incorrect expectations for these unusual sentences, and the results do not reflect any pure SUMMARY effect of syntactic complexity on comprehension (Gennari and The two cases reviewed here, verb modification ambiguities and MacDonald, 2008) . relative clauses, exemplify the PDC’s point that an understanding of production choices in a language is critical for understand- ing comprehension. That idea has been implicit in non-syntactic A fuller treatment than is presented here would include the fact that object relatives with pronoun embedded subjects (The boy/toy she splashed....) comprehension work for decades (e.g., in lexical frequency effects have different production biases, different rates of production, and differ- on word recognition, in that frequency is inherently an effect ent comprehension patterns than the examples discussed here. We must of experience and ultimately producers’ word choices), but it’s also consider whether Easy First, Plan Reuse, and Reduce Interference pro- vide an adequate account of why multiply-embedded object relatives, like Miller and Chomsky’s (1963) The rat [the cat [the dog chased] killed] ate the difficulty here can also be traced to ambiguity resolution gone awry rather malt, are essentially never produced, and the extent to which comprehension than hard limits on working memory capacity. Frontiers in Psychology | Language Sciences April 2013 | Volume 4 | Article 226 | 12 MacDonald Language production shapes form and comprehension quite another thing to claim that we could abandon many of the This linkage between action planning and the mechanisms special-purpose syntactic interpretation mechanisms common in of language production has several intriguing implications the parsing literature if we understood sentence production bet- for the way language researchers view language form and ter. It will take some time to test this view in other constructions use. First, an implication for psycholinguistics: For decades, and languages, but in the meantime, the availability of exten- experience-based sentence comprehension research has empha- sive language corpora in many languages permits comprehension sized the non-independence of lexical and syntactic represen- researchers to examine the relationship between production pat- tations during the comprehension process (e.g., MacDonald terns (in the corpus) and comprehension behavior, even if they et al., 1994). By contrast, language production and motor/action have not yet investigated the production pressures that create planning more generally rely on abstract high-level plans the distributional regularities that are observed in a corpus. The that appear quite independent from the elements in the PDC suggests that it is essential to investigate such linkages before plan. Understanding how the demands of comprehension declaring that comprehension behavior owes to highly specific and production integrate lexical and more abstract hierarchi- design features in the language comprehension system. cal representations is an important challenge as these fields move forward. One possibility is that comprehension pro- IMPLICATIONS, LIMITATIONS, FUTURE DIRECTIONS cesses may draw on covert language production processes and The PDC begins with something utterly uncontroversial, that other aspects of non-linguistic motor planning (Pickering and language production is hard. The next step is no less obvious Garrod, 2007). If so, production may be doubly intertwined to production researchers, that language producers try to make with comprehension, both in the PDC’s view of produc- things easier, and that their attempts affect the form of the utter- tion mechanisms generating the statistics of language forms ances they produce. From there we get into somewhat more that drive comprehension, and also Pickering and Garrod’s controversial territory that (a) producers’ choices of utterance argument for covert production processes in the service of forms, repeated through the population, have a significant role in comprehension. explaining language typology and change over time, and (b) lan- Second, the link between language and action planning has guage users learn these statistical patterns and rapidly use them implications for how we view language itself. An enormous lit- to interpret new input. There are aspects of all of these ideas in erature considers how language is distinct from non-linguistic the literature, but the PDC is greater than the sum of these parts cognition (see Newmeyer, 1998; Jackendoff, 2002, 2007,among in suggesting that the downstream influences of production pro- numerous others), but the PDC may be able to contribute cesses are so strong and so pervasive that we must take production to the discussion. There has been little work to date investi- processes into account in developing theories of language form, gating the commonalities and differences between the abstract change, and comprehension. hierarchical plans that underlie sentence production and those One of the ways that the PDC is different from related ideas that underlie non-linguistic motor behavior. To the extent that is its emphasis on a specifically mechanistic account of language such commonalities exist, they could suggest that syntax, at production. It is certainly not wrong to appeal to more abstract least as it is realized in creating utterances, has a potential notions of communicative efficiency in accounting for producers’ homologue in non-linguistic systems and therefore is not some- choices of utterance forms (e.g., Jaeger and Tily, 2011; Piantadosi thing that distinguishes language from other cognition. However, et al., 2012), but the PDC can offer something more to the extent linguistic utterances clearly differ from other actions in that that it draws on the mechanisms of memory retrieval, attention, they have both a goal (e.g., to communicate) and a mean- serial order maintenance, and motor planning in understanding ing, while complex actions have a goal (e.g., to make coffee), what is more vs. less efficient. Similarly, Bybee (2006) and oth- and a hierarchical plan to realize the goal, but no inherent ers make important claims that language use, broadly construed, meaning. This meaning and its interplay with utterance form, underlies language typology and change, but the PDC aims to meted out over time as the language is planned, produced, be more specific: Language producibility, more than learnabil- and comprehended, would seem to be a critical aspect of what ity or comprehensibility, drives language form. The reasons for makes language unlike non-linguistic cognition. Again, work this claim again invoke mechanistic accounts of language pro- toward a mechanistic account of how language is planned and duction to explain what is difficult, how producers manage that uttered may have consequences well beyond the field of language difficulty, and how they are the primary controllers of utter- production. ance form. There’s a great deal of work remaining in order to realize this goal of a mechanistic account of language pro- ACKNOWLEDGMENTS duction, including extensions beyond the lexico-syntactic focus Preparation of this article was supported by grants from the of this article. Working toward a more mechanistic account National Institutes of Health (R01 HD047425), the National is important because links to memory, action planning, and Science Foundation (BCS 1123788), and the Wisconsin Alumni other non-linguistic domains can ground the PDC approach Research Fund. This work greatly benefitted from comments in broader cognitive processes and avoid potential circulari- by Gary Lupyan, Jenny Saffran, Mark Seidenberg, and mem- ties among what is efficient, common, easy, salient, and other bers of the Language and Cognitive Neuroscience Lab at constructs that are invoked in many accounts of language form the University of Wisconsin-Madison. Errors are of course and use. my own. www.frontiersin.org April 2013| Volume4| Article226 | 13 MacDonald Language production shapes form and comprehension Bresnan, J., and Ford, M. (2010). Cooper, R., and Shallice, T. (2000). Ferreira, V. S., and Dell, G. S. (2000). REFERENCES Acheson, D. J., and MacDonald, Predicting syntax: processing dative Contention scheduling and the Effect of ambiguity and lexical M. C. (2009). Verbal working constructions in American and control of routine activities. Cogn. availability on syntactic and lexi- memory and language produc- Australian varieties of English. Neuropsychol. 17, 297–338. cal production. Cognit. Psychol. 40, tion: common approaches to Language 86, 168–213. Cowan, N. (2005). Working Memory 296–340. the serial ordering of verbal Bybee, J. (2006). From usage to gram- Capacity. New York, NY: Taylor and Ferrera, V. P. (2000). Task-dependent information. Psychol. Bull. 135, mar: the mind’s response to repeti- Francis. modulation of the sensorimotor 50–68. tion. Language 82, 711–733. Culbertson, J., Smolensky, P., and transformation for smooth pursuit Allen, J., and Seidenberg, M. S. (1999). Bybee, J., and McClelland, J. L. (2005). Legendre, G. (2012). Learning eye movements. J. Neurophysiol. 84, “The emergence of grammaticality Alternatives to the combinatorial biases predict a word order 2725–2738. in connectionist networks,” in paradigm of linguistic theory based universal. Cognition 122, 306–329. Fidelholtz, J. (1975). Word frequency The Emergence of Language, ed on domain general principles of Dahan, D., Tanenhaus, M. K., and and vowel reduction in English. B. MacWhinney (Mahwah, NJ: human cognition. Linguist. Rev. 22, Chambers, C. G. (2002). Accent Chic. Linguist. Soc. 11, 200–213. Erlbaum), 115–151. 381–410. and reference resolution in spoken- Fodor, J. D., and Inoue, A. (1994). The Allen, K., Ibara, S., Seymour, A., Carreiras, M., Duñabeitia, J. A., language comprehension. J. Mem. diagnosis and cure of garden paths. Cordova, N., and Botvinick, M. Vergara, M., de la Cruz-Pavía, I., Lang. 47, 292–314. J. Psycholinguist. Res. 23, 407–434. (2010). Abstract structural repre- and Laka, I. (2010). Subject relative Dakin, S.C., Mareschal, I., andBex,P. Fox Tree, J. E., and Clark, H. H. (1997). sentations of goal-directed behavior. clauses are not universally easier J. (2005). An oblique effect for local Pronouncing “the” as “thee” to sig- Psychol. Sci. 21, 1518–1524. to process: evidence from Basque. motion: psychophysics and natural nal problems in speaking. Cognition Altmann, G. T. M.,van Nice,K.Y., Cognition 115, 79–92. movie statistics. J. Vis. 5, 878–887. 62, 151–167. Garnham, A., and Henstra, J.-A. Chafe, W. (1976). “Givenness, con- Deco, G., and Rolls, E. T. (2005). Frazier, L. (1987). “Sentence process- (1998). Late closure in context. trastiveness, definiteness, subjects, Sequential memory: a putative ing: a tutorial review,” in Attention J. Mem. Lang. 38, 459–484. topics, and point of view,” in neural and synaptic dynamical and Performance, Vol. 12. The psy- Anderson, J. R. (1983). Retrieval of Subject and Topic, ed C. Li mechanism. J. Cogn. Neurosci. 17, chology of reading, ed M. Coltheart information from long-term mem- (New York, NY: Academic Press), 294–307. (Hove: Erlbaum). ory. Science 220, 25–30. 25–55. Dell, G. S., Burger, L. K., and Svec, Frazier, L. (1995). Constraint satisfac- Aslin, R. N., and Newport, E. L. (2012). Chang, F. (2009). Learning to order W. R. (1997). Language production tion as a theory of sentence pro- Statistical learning: from acquiring words: a connectionist model of and serial order: a functional anal- cessing. J. Psycholinguist. Res. 24, specific items to forming general heavy NP shift and accessibility ysis and a model. Psychol. Rev. 104, 437–468. rules. Curr.Dir.Psychol.Sci. 21, effects in Japanese and English. 123–147. Gennari,S.P., and MacDonald,M. C. 170–176. J. Mem. Lang. 61, 374–397. Dikker, S., Rabagliati, H., Farmer, T. (2008). Semantic indeterminacy in Babyonyshev, M., and Gibson, E. Chang, F., Dell, G. S., and Bock, K. A., and Pylkkänen, L. (2010). Early object relative clauses. J. Mem. Lang. (1999). The complexity of nested (2006). Becoming syntactic. Psychol. occipital sensitivity to syntactic cat- 58, 161–187. structures in Japanese. Language 75, Rev. 113, 234–272. egory is based on form typicality. Gennari, S. P., and MacDonald, M. 423–450. Chomsky, N. (1957). Syntactic Psychol. Sci. 21, 629–634. C. (2009). Linking production and Bard, E. G., Anderson, A. H., Chen, Structures. The Hague: Mouton. Duran, N. D., Dale, R., and Kreuz, comprehension processes: the case Y.,Nicholson, H. B. M.,Havard, C., Chomsky, N., and Miller, G. A. (1963). R. J. (2011). Listeners invest in an of relative clauses. Cognition 111, and Dalzel-Job, S. (2007). Let’s you “Introduction to the formal analysis assumed other’s perspective despite 1–23. do that: sharing the cognitive bur- of natural languages,” in Handbook cognitive cost. Cognition 121, 22–40. Gennari,S.P., Mirkovic, ´ J., and dens of dialogue. J. Mem. Lang. 57, of Mathematical Psychology,Vol.2, Echols, C. H., and Newport, E. L. MacDonald, M. C. (2012). Animacy 616–641. eds R. D. Luce, R. R. Bush, and (1992). The role of stress and posi- and competition in relative clause Bock, J. K. (1982). Toward a cogni- E. Galanter (New York, NY: Wiley), tion in determining first words. production: a cross-linguistic tive psychology of syntax: infor- 269–321. Lang. Acquis. 2, 189–220. investigation. Cognit. Psychol. 65, mation processing contributions to Christiansen,M.H., andChater, Evans, N., and Levinson, S. C. (2009). 141–176. sentence formulation. Psychol. Rev. N. (2001). “Finite models of The myth of language universals: Gibson, B. M., Wasserman, E. A., 89, 1–47. infinite language: a connection- language diversity and its impor- and Kamil, A. C. (2007). Pigeons Bock, J. K., and Warren, R. K. (1985). ist approach to recursion,” in tance for cognitive science. Behav. and people select efficient routes Conceptual accessibility and syntac- Connectionist Psycholinguistics, eds Brain Sci. 32, 429–448. when solving a one-way “trav- tic structure in sentence formula- M. H. Christiansen and N. Chater Feist, M. I. (2010). “Inside in and eling salesperson” task. J. Exp. tion. Cognition 21, 47–67. (Westport, CT: Ablex), 138–176. on: typological and psycholin- Psychol. Anim. Behav. Process. 33, Botvinick, M., and Plaut, D. C. (2004). Christianson, K., and Ferreira, F. guistic perspectives,” in Language, 244–261. Doing without schema hierarchies: (2005). Conceptual accessibility and Cognition, and Space, eds V. Evans Gibson, E. (1998). Linguistic complex- a recurrent connectionist approach sentence production in a free word and P. Chilton (London: Equinox). ity: locality of syntactic dependen- to normal and impaired routine order language (Odawa). Cognition Ferreira, F. (1991). Effects of length cies. Cognition 68, 1–76. sequential action. Psychol. Rev. 111, 98, 105–135. and syntactic complexity on initia- Gibson, E., Pearlmutter, N., Canseco- 395–429. Clahsen, H., Lück, M., and Hahne, tion times for prepared utterances. Gonzalez, E., and Hickok, G. Branigan, H. P., Pickering, M. J., and A. (2007). How children process J. Mem. Lang. 30, 210–233. (1996). Recency preference in Cleland, A. A. (1999). Syntactic over-regularizations: evidence from Ferreira, F. (1994). Choice of passive the human sentence processing priming in written production: evi- event-related brain potentials. voice is affected by verb type and mechanism. Cognition 59, 23–59. dence for rapid decay. Psychon. Bull. J. Child Lang. 34, 601–622. animacy. J. Mem. Lang. 33, 715–736. Givón, T. (1985). “Function, struc- Rev. 6, 635–640. Connine,C. M., andClifton,C. J. Ferreira, F., and Swets, B. (2002). How ture, and language acquisition,” Bresnan, J., Cueni, A., Nikitina, T., and (1987). Interactive use of lexi- incremental is language production? in The Crosslinguistic Study of Baayen, R. H. (2007). “Predicting cal information in speech percep- Evidence from the production of Language Acquisition,Vol.2,ed D. the dative alternation,” in Cognitive tion. J. Exp. Psychol. Hum. Percept. utterances requiring the computa- Slobin. (Hillsdale, NJ: Erlbaum), Foundations of Interpretation, Perform. 13, 291–299. tion of arithmetic sums. J. Mem. 1005–1028. eds G. Boume, I. Kraemer, and Conrad, R., and Hull, A. J. (1964). Lang. 46, 57–84. Goldberg, A. (2006). Constructions at J. Zwarts (Amsterdam: Royal Information, acoustic confusion Ferreira, V. S., and Bock, K. (2006). Work: The Nature of Generalization Netherlands Academy of Science), and memory span. Br.J.Psychol. 55, The functions of structural priming. in Language. Oxford: Oxford 69–94. 429–432. Lang. Cogn. Process. 21, 1011–1029. University Press. Frontiers in Psychology | Language Sciences April 2013 | Volume 4 | Article 226 | 14 MacDonald Language production shapes form and comprehension Grodner, D., and Gibson, E. (2005). (2011). Tracking talking: dual task sentence comprehension,” in visually situated sentence pro- Consequences of the serial nature of costs of planning and produc- Handbook of Psycholinguistics, eds duction. Front. Psychol. 2:396. doi: linguistic input for sentenial com- ing speech for young versus older M. Traxler and M. A. Gernsbacher 10.3389/fpsyg.2011.00396 plexity. Cogn. Sci. 29, 261–290. adults. Aging Neuropsychol. Cogn. (New York, NY: Academic Press), Newmeyer, F. J. (1998). Language Form Grossberg, S. (1978). A theory of 18, 257–279. 581–611. and Language Function. Cambridge, human memory: self-organization Kimball, J. (1973). Seven principles of MacDonald, M. C., and Thornton, R. MA: MIT Press. and performance of sensory-motor surface structure parsing in natural (2009). When language comprehen- O’Brien, J. L., and Raymond, J. E. codes, maps, and plans. Prog. Theor. language. Cognition 2, 15–47. sion reflects production constraints: (2012). Learned predictiveness Biol. 5, 233–374. Kraljic, T., Samuel, A. G., and Brennan, resolving ambiguities with the help speeds visual processing. Psychol. Hale, J. (2006). Uncertainty about the S. E. (2008). First impressions and of past experience. Mem. Cognit. 37, Sci. 23, 359–363. rest of the sentence. Cogn. Sci. 30, last resorts: how listeners adjust to 1177–1186. O’Grady, W. (2011). “Relative clauses: 643. speaker variability. Psychol. Sci. 19, Mack,J.E., Clifton, C., Frazier,L., and processing and acquisition,” in Hart, B., and Risley, T. R. (1995). 332–338. Taylor, P. V. (2012). (Not) Hearing The Acquisition of Relative Clauses: Meaningful Differences in the Kuchinsky, S. E., Bock, K., and Irwin, optional subjects: the effects of Processing, Typology and Function, Everyday Experience of Young D. E. (2011). Reversing the hands pragmatic usage preferences. ed E. Kidd (Amsterdam: John American Children.Baltimore, MD: of time: changing the mapping from J. Mem. Lang. 67, 211–223. Benjamins), 13–38. Brookes. seeing to saying. J. Exp. Psychol. MacNeilage, P.F., and Davis,B.L. Piantadosi, S. T., Tily, H., and Gibson, Hartley, T., and Houghton, G. (1996). A Learn. Mem. Cogn. 37, 748–756. (2000). On the origin of internal E. (2012). The communicative func- linguistically constrained model of Kuperman, V., and Bresnan, J. (2012). structure of word forms. Science tion of ambiguity in language. short-term memory for nonwords. The effects of construction prob- 288, 527–531. Cognition 122, 280–291. J. Mem. Lang. 35, 1–31. ability on word durations dur- Miller, G. A. (1956). The magical num- Pickering, M. J., and Garrod, S. (2007). Hawkins, J. A. (2004). Efficiency and ing spontaneous incremental sen- ber seven, plus or minus two: some Do people use language production Complexity in Grammars. Oxford: tence production. J. Mem. Lang. 66, limits on our capacity for process- to make predictions during com- Oxford University Press. 588–611. ing information. Psychol. Rev. 63, prehension? Trends Cogn. Sci. 11, Horton, W. S., and Keysar, B. (1996). Lashley, K. S. (1951). “The problem of 81–97. 105–110. When do speakers take into account serial order in behavior,” in Cerebral Miller, G. A., and Chomsky, N. (1963). Postle, B. R. (2006). Working memory common ground? Cognition 59, Mechanisms in Behavior; the Hixon “Finitary models of language users,” as an emergent property of the mind 91–117. Symposium, ed L. A. Jeffress (Oxford in Handbook of Mathematical and brain. Neuroscience 139, 23–38. Hudson, R. (1996). The difficulty England: Wiley), 112–146. Psychology,Vol.2, eds R. D. Luce, Ramscar, M., and Yarlett, D. (2007). of (so-called) self-embedded Levelt, W. J. M. (1982). “Linearization R. R. Bush, and E. Galanter. (New Linguistic self-correction in the structures. Work.Pap.Linguist. 8, in describing spatial networks,” in York, NY: Wiley), 419–491. absence of feedback: a new 283–314. Processes, Beliefs, and Questions, Miller, G. A., and Selfridge, J. A. (1950). approach to the logical prob- Hudson Kam, C. L., and Newport, E. eds S. Peters and E. Saarinen Verbal context and the recall of lem of language acquisition. Cogn. L. (2009). Getting it right by getting (Dordrecht: Reidel), 199–220. meaningful material. Am. J. Psychol. Sci. 31, 927–960. it wrong: when learners change lan- Levy, R. (2008). Expectation-based 63, 176–185. Reali, F., and Christiansen, M. H. guages. Cognit. Psychol. 59, 30–66. syntactic comprehension. Cognition Mitchell, D. C., and Brysbaert, M. (2007). Word chunk frequencies Jackendoff, R. (2002). Foundations 106, 1126–1177. (1998). “Challenges to recent the- affect the processing of pronomi- of Language: Brain, Meaning, Lightfoot, D. (1999). The Development ories of cross-linguistic differences nal object-relative clauses. Q. J. Exp. Grammar, Evolution. Oxford: of Language: Acquisition, Change, in parsing: evidence from Dutch,” Psychol. 60, 161–170. Oxford University Press. and Evolution. Oxford: Blackwell. in Sentence Processing: A Cross- Rokem, A., and Silver, M. A. (2009). Jackendoff, R. (2007). Linguistics in Lin, C. J. C. (2008). The process- Linguistic Perspective, ed D. Hillert A model of encoding and decod- cognitive science: the state of the art, ing foundation of head-final relative (San Diego, CA: Academic Press), ing in V1 and MT accounts for Linguist. Rev. 24, 347–401. clauses. Lang. Linguist. 9, 813–838. 313–335. motion perception anisotropies in Jaeger, T. F., and Tily, H. (2011). On lan- MacDonald, M. C. (1994). Probabilistic Miyata, H., and Fujita, K. (2011). the human visual system. Brain Res. guage “utility”: processing complex- constraints and syntactic ambiguity Flexible route selection by pigeons 1299, 3–16. ity and communicative efficiency. resolution. Lang. Cogn. Process. 9, (Columba livia)onacomputer- Rosenbaum, D. A., Cohen, R. G., Jax, S. WIREs Cogn. Sci. 2, 323–335. 157–201. ized multi-goal navigation task with A.,Weiss, D. J.,and vander Wel, R. Jäger, G., and Rosenbach, A. (2008). MacDonald, M. C. (1999). and without an “obstacle.” J. Comp. (2007). The problem of serial order Priming and unidirectional lan- “Distributional information in Psychol. 125, 431–435. in behavior: Lashley’s legacy. Hum. guage change. Theor. Linguist. 34, language comprehension, produc- Montag, J. L., and MacDonald, M. Mov. Sci. 26, 525–554. 85–113. tion, and acquisition: three puzzles C. (2009). “Word order doesn’t Roth, F. P. (1984). Accelerating lan- Jin, D. Z. (2009). Generating vari- and a moral,” in The Emergence matter: relative clause produc- guage learning in young children. able birdsong syllable sequences of Language, ed B. MacWhinney tion in English and Japanese,” in J. Child Lang. 11, 89–107. with branching chain networks (Mahwah, NJ: Lawrence Erlbaum Proceedings of the 31th Annual Rumelhart, D. E., and McClelland, J. in avian premotor nucleus HVC. Associates Publishers), 177–196. Conference of the Cognitive Science L. (1986). “On learning the past Phys. Rev. E. Stat. Nonlin. Soft MacDonald, M. C., and Christiansen, Society, eds N. A. Taatgen and H. tense of English verbs,” in Parallel Matter Phys. 80(5 Pt 1):051902. doi: M. H. (2002). Reassessing work- van Rijn (Austin, TX: Cognitive Distributed Processing: Explorations 10.1103/PhysRevE.80.051902 ing memory: comment on Just and Science Society), 2594–2599. in the Microstructure of Cognition, Kaiser, E. (2012). Taking action: Carpenter (1992) and Waters and Müller, H., and Sternad, D. (2004). Vol. 2, eds J. L. McClelland, a cross-modal investigation of Caplan (1996). Psychol. Rev. 109, Decomposition of variability in D. E. Rumelhart and the PDP discourse-level representations. 35–54. the execution of goal-oriented Research Group. (Cambridge, Front. Psychol. 3:156. doi: 10.3389/ MacDonald, M. C., Pearlmutter, N. tasks: three components of skill MA: MIT Press/Bradford Books), fpsyg.2012.00156 J., and Seidenberg, M. S. (1994). improvement. J. Exp. Psychol. 217–271. Karpicke, J. D. (2012). Retrieval-based The lexical nature of syntactic Hum. Percept. Perform. 30, Scott-Phillips, T. C., and Kirby, S. learning: active retrieval promotes ambiguity resolution. Psychol. Rev. 212–233. (2010). Language evolution in the meaningful learning. Curr. Dir. 101, 676–703. Myachykov, A., Thompson, D., laboratory. Trends Cogn. Sci. 14, Psychol. Sci. 21, 157–163. MacDonald, M. C., and Seidenberg, Garrod, S., and Scheepers, C. 411–417. Kemper, S., Hoffman, L., Schmalzried, M. S. (2006). “Constraint satis- (2011). Referential and visual Seidenberg, M. S. (1997). Language R., Herman, R., and Kieweg, D. faction accounts of lexical and cues to structural choice in acquisition and use: learning and www.frontiersin.org April 2013| Volume4| Article226 | 15 MacDonald Language production shapes form and comprehension applying probabilistic constraints. first-pass structural attachment abandoned context words affect a role for (co-)articulatory fluency? Science 275, 1599–1604. preferences with dynamic speech production. Acta Psychol. J. Mem. Lang. 58, 48–65. Seidenberg, M. S., and Zevin, J. D. grammars and recursive neural 140, 218–229. Yamashita, H., and Chang, F. (2001). (2006). “Connectionist models in networks. Cognition 88, 133–169. Van Petten, C., and Luka, B. J. (2012). “Long before short” preference developmental cognitive neuro- Tanaka, M. N., Branigan, H. P., Prediction during language compre- in the production of a head- science: critical periods and the McLean, J. F., and Pickering, hension: benefits, costs, and ERP final language. Cognition 81, paradox of success,” in Attention M. J. (2011). Conceptual influ- components. Int. J. Psychophysiol. B45–B55. and Performance XXI: Processes ences on word order and voice 83, 176–190. of Change in Brain and Cognitive in sentence production: evidence Wanner, E., and Maratsos, M. (1978). Conflict of Interest Statement: The Development, eds Y. Munakata from Japanese. J. Mem. Lang. 65, “An ATN approach to compre- authors declare that the research and M. Johnson (Oxford: Oxford 318–330. hension,” in Linguistic Theory was conducted in the absence of any University Press), 585–612. Tarr, M. J., and Gauthier, I. (2000). and Psychological Reality, eds M. commercial or financial relationships Singleton, J. L., and Newport, E. L. FFA: a flexible fusiform area for Halle, J. Bresnan, and G. A. Miller. that could be construed as a potential (2004). When learners surpass subordinate-level visual processing (Cambridge, MA US: MIT Press), conflict of interest. their models: The acquisition of automatized by expertise. Nat. 119–161. American Sign Language from Neurosci. 3, 764–769. Warren, R. M., and Sherman, G. L. Received: 08 November 2012; paper inconsistent input. Cognit. Psychol. Thornton, R., MacDonald, M. C., and (1974). Phonemic restorations pending published: 25 November 2012; 49, 370–407. Gil, M. (1999). Pragmatic con- based on subsequent context. accepted: 11 April 2013; published Smith, M., and Wheeldon, L. (2004). straint on the interpretation of com- Percept. Psychophys. 16, 150–156. online: 26 April 2013. Horizontal information flow in spo- plex noun phrases in Spanish and Wasow, T. (1997). Remarks on gram- Citation: MacDonald MC (2013) How ken sentence production. J. Exp. English. J. Exp. Psychol. Learn. Mem. matical weight. Lang. Variation language production shapes language Psychol. Learn. Mem. Cogn. 30, Cogn. 25, 1347–1365. Change 9, 81–105. form and comprehension. Front. Psychol. 675–686. Traxler, M. J., Morris, R. K., and Seely, Weiner, E. J., and Labov, W. (1983). 4:226. doi: 10.3389/fpsyg.2013.00226 Stallings, L. M., MacDonald, M. C., R. E. (2002). Processing subject Constraints on the agentless passive. This article was submitted to Frontiers and O’Seaghdha, P. G. (1998). and object relative clauses: evidence J. Linguist. 19, 29–58. in Language Sciences, a specialty of Phrasal ordering constraints in from eye movements. J. Mem. Lang. Wells, J. B., Christiansen, M. H., Frontiers in Psychology. sentence production: Phrase 47, 69–90. Race, D. S., Acheson, D. J., Copyright © 2013 MacDonald. This is length and verb disposition in Tse, C.-S., Li, Y., and Altarriba, J. and MacDonald, M. C. (2009). an open-access article distributed under heavy-NP shift. J. Mem. Lang. 39, (2011). The effect of semantic relat- Experience and sentence process- the terms of the Creative Commons 392–417. edness on immediate serial recall ing: statistical learning and relative Attribution License,which permits use, Stemberger, J. P. (1989). Speech errors and serial recognition. Q. J. Exp. clause comprehension. Cognit. distribution and reproduction in other in early child language production. Psychol. 64, 2425–2437. Psychol. 58, 250–271. forums, provided the original authors J. Mem. Lang. 28, 164–188. Tydgat, I., Diependaele, K., Hartsuiker, Woodward, A. J., Macken, W. J., and and source are credited and subject to any Sturt, P., Costa, F., Lombardo, V., R. J., and Pickering, M. J. (2012). Jones, D. M. (2008). Linguistic copyright notices concerning any third- and Frasconi, P. (2003). Learning How lingering representations of familiarity in short-term memory: party graphics etc. Frontiers in Psychology | Language Sciences April 2013 | Volume 4 | Article 226 | 16

Journal

Frontiers in PsychologyPubmed Central

Published: Apr 26, 2013

References