Elsevier

Cognition

Volume 111, Issue 1, April 2009, Pages 1-23
Cognition

Linking production and comprehension processes: The case of relative clauses

https://doi.org/10.1016/j.cognition.2008.12.006Get rights and content

Abstract

Six studies investigated the relationship between production and comprehension by examining how relative clause production mechanisms influence the probabilistic information used by comprehenders to understand these structures. Two production experiments show that accessibility-based mechanisms that are influenced by noun animacy and verb type shape relative clause production. Two corpus studies confirm these production mechanisms in naturally occurring productions. Two comprehension studies found that nouns and verb types occurring in structures that speakers do not produce are difficult to comprehend. Specifically, the probability of producing a passive structure for a verb type in a given animacy configuration, as measured in the production and corpus studies, predicts comprehension difficulty in active structures. Results suggest that the way in which the verb roles are typically mapped onto syntactic arguments in production plays a role in comprehension. Implications for the relationship between production, comprehension and language learning are discussed.

Introduction

One of the most central questions in cognitive science is the extent to which humans’ information processing abilities are shaped by learning from experience. While researchers in every field agree that learning plays a role, there is significant disagreement concerning the extent of this influence. For example, as evidenced by both behavioral and imaging data (Yovel & Kanwisher, 2004), human face recognition processes are distinctly different in character from the processes that recognize natural and manufactured objects, such as trees and cups. One interpretation of these results is that the striking differences between face and object processing reflect the operations of a face-specific processing system that has primarily been shaped by evolutionary forces (Grill-Spector et al., 2004, McKone et al., 2007). An alternative explanation is that these results reflect humans’ learning from their unique experiences with faces (Gauthier et al., 2003, Tarr and Gauthier, 2000). On this view, the face processing data are different in character from other object recognition data because humans have more and different experiences with faces compared to other entities in the world. By virtue of learning from these experiences and unique task demands (such as identifying individuals rather than categorizing them), humans’ face processing has a unique character. This view places the burden of the explanation for face processing on learning from experience: the distributional regularities in the input (the range and frequency of faces in the visual world), the special task demands for faces, and the learning abilities of the perceiver to adapt to these experiences and demands.

Similar debates pervade research in language acquisition and processing. Within language acquisition, there are many proposals for a major role of innate grammatical constraints or cognitive primitives, which assign a peripheral role to linguistic experience (Chomsky, 1959, Lust and Foley, 2004), and there are also advocates for ascribing a large role to learning processes (Tomasello, 2003). Similarly, within adult language comprehension, some accounts posit a sentence processing mechanism that operates by fixed parsing principles that navigate the pervasive ambiguities in language (Frazier, 1987), while other traditions have argued that comprehension processes weigh the probability of alternative interpretations based on learning from prior experience (MacDonald et al., 1994, Tanenhaus and Trueswell, 1995). These debates in the language domain have an additional level of complexity that is not present in the face processing example, which is that human language users are both potential learners from their linguistic experience and the creators of that experience. That is, the distributional regularities of faces stem from human physiology, not so much human behavior, but the language statistics that comprehenders learn originate in utterances that they and other language users produce.

Whereas researchers in face perception do not feel compelled to explain why faces have the particular distributional characters they have, language researchers can investigate why utterances have certain properties and not others. Indeed, some have argued that an account of language comprehension as emerging from language statistics is incomplete without an account of where the statistics come from, i.e., why languages have particular distributional characteristics and not others (Frazier, 1995; but cf. MacDonald, 1997). Our goal in this article is to take steps in developing a unified framework that explains both the origin of the distributional regularities of language and how learning of those regularities shapes the language comprehension system. The question concerning the origin of linguistic patterns is an important area of research in historical linguistics (Hopper and Traugott, 1993, Keller, 1994, Sweetser, 1990) and to a lesser extent in psycholinguistics (Hawkins, 2004, Wasow, 2002). Although there are multiple reasons why languages tend to have certain properties, our claim is that language production processes are a significant influence on the distributional patterns to which a language user is exposed.

Our approach is called the Production-Distribution-Comprehension (PDC) framework because it attempts to link properties of the language production system to particular choices made during utterance production, to link those choices to particular distributional patterns in the input provided to comprehenders, and finally to show that comprehension behavior is modulated by these distributional patterns (MacDonald, 1999, MacDonald and Thornton, unpublished manuscript). At the level of sentence processing, the PDC account essentially argues that structure choices in production, at least some of which are determined by production-specific mechanisms, create robust distributional patterns in the language, which are learned over time by comprehenders who are exposed to this input. These distributional patterns then become the probabilistic constraints that guide the comprehension process in a constraint-based system. On this view, comprehension processes and interpretation preferences can thus be traced to distributional patterns in language use, which are themselves emergent from production mechanisms affecting speakers and writers.

While language producers clearly do make some production choices for the benefit of their audience (e.g., Clark, 1996), it is also clear that some production choices emerge because of the needs of the speaker (Ferreira & Dell, 2000). During planning, choice of word order and syntactic structure is strongly constrained by the accessibility of words and phrases (e.g., Bock, 1982, Bock, 1987, Bock and Irwin, 1980, Bock and Levelt, 1994, Bock and Loebell, 1990, Bock and Warren, 1985, McDonald et al., 1993). Accessibility can be understood as the degree to which a word or phrase is ready for articulation in the utterance – some elements, by virtue of being long, rare, less frequent or conceptually less salient, etc., may require more planning and retrieval time than others. Given the incremental nature of production in which uttering and planning of upcoming elements occurs simultaneously (Ferreira & Swets, 2002), fluency is maximized by uttering more accessible portions of an utterance early, leaving additional time to plan less accessible components. Our claim is that accessibility-based and other production-driven choices yield distributional patterns in the language that ultimately shape comprehension processes.

Investigating a claim of this sort has several steps, including identifying what production choices speakers make, testing whether the production processes involved give rise to broad distributional patterns, and determining the extent to which the distributional patterns predict comprehension performance. In this article, we follow these steps with reference to the production and comprehension of object relative (or center embedded) clauses such as the lawyer that the judge criticized. Dating from Miller and Chomsky’s (1963) first discussion of center embedding, an extensive literature has been devoted to explaining why these structures are difficult for children and adults to comprehend (e.g. Kidd et al., 2007, Sheldon, 1974, Wanner and Maratsos, 1978, Caplan and Waters, 1999, Mak et al., 2002, Mak et al., 2006). Object relative clauses therefore form a well-studied domain in which to investigate links between production and adult processing, and eventually, links to acquisition. In the present article, we investigate speakers’ choices in relative clause production, the resulting distributional regularities in input to comprehenders, and the consequences of these regularities on comprehension processes. We first briefly review the large literature on relative clause comprehension before turning to the much less studied processes of relative clause production.

Object relative clauses have been found to be more difficult to comprehend than subject relative clauses (e.g., the lawyer that criticized the judge), which contain the same words but a different word order and meaning, or passive relative clauses (e.g., the lawyer that was criticized by the judge), which have a similar meaning (Waters and Caplan, 1996, King and Just, 1991, Just and Carpenter, 1992, Gordon et al., 2001, Gordon et al., 2004, Traxler et al., 2002, Mak et al., 2002, MacWhinney and Pleh, 1988, Gennari and MacDonald, 2008). These findings have often been thought to stem from architectural properties of the sentence comprehension processes, for example that object relatives are harder because they tax working memory more than do subject relatives (see Gennari & MacDonald, 2008, for review). A more recent approach points to the role of learning of probabilistic regularities in explaining comprehension difficulty and rate of acquisition, particularly properties of the nouns in the relative clause (Gennari and MacDonald, 2008, Kidd et al., 2007, Mak et al., 2002, Reali and Christiansen, 2007, Wells et al., 2009). Several studies have pointed out that comprehension difficulty in object relatives varies with the animacy configuration of the nouns involved (Mak et al., 2002, Traxler et al., 2002): Object relatives with inanimate heads such as that in (1a) are easier to process than those with animate heads in (1b).

  • (1a)

    The movie that the director watched received a prize.

  • (1b)

    The director that the movie pleased received a prize.

This animacy effect matches the frequency of the animacy configurations in several corpora: inanimate-head object relative clauses are more frequent than animate-head ones (Roland et al., 2007, Mak et al., 2002). Gennari and MacDonald (2008) explicitly linked noun animacy to the availability of alternative interpretations in object relatives. They showed that word-by-word reading difficulty in these structures correlates with the availability of the alternative interpretations as the clause unfolds, as measured by sentence completion studies. They argued, along the lines of the constraint satisfaction approach, that semantic indeterminacy plays a role in explaining comprehension difficulty, as noun animacy provides probabilistic information modulating the relative likelihood of the thematic roles assigned to the nouns and the relative likelihood of the alternative events into which the nouns can be integrated.

In the present work, we extend this research to production processes themselves, focusing on the role of animacy and verb properties, such as their associated mappings from event roles to syntactic arguments. Specifically, we argue that constraints during utterance planning give rise to production choices in which certain verbs and noun types co-vary with a particular choice of active or passive structure within the relative clause, resulting in a particular mapping from event roles to syntactic arguments. In active structures, for example, agents or instigators of the event tend to be mapped onto subject position, but this is not the case for passive constructions. The particular conjunctions of verb type and animacy configurations (that is, the animate vs. inanimate status of subject and object nouns) that tend to be produced in passive structures are necessarily rare in active structures. Comprehenders are therefore misled when encountering an active object relative with an animacy configuration and a verb type that signal a passive structure. The unfolding animacy configuration generates great semantic indeterminacy as to the roles that the nouns may play relative to the verb (Gennari & MacDonald, 2008). As proposed by constraint satisfaction approaches, competition between alternative thematic role interpretations of the nouns and their co-varying passive and active structures ensues, as dictated by language experience, giving rise to comprehension difficulty (see, for example, the competition-integration model of McRae, Spivey-Knowlton, & Tanenhaus, 1998).1 Comprehension difficulty and interpretation preferences in object relatives can thus be traced to frequency-guided interpretation preferences, themselves traced to distributional patterns derived from production mechanisms, which promote certain choices among speakers and thus the distributional patterns that drive the comprehension process.

A number of syntactic choices in production have been shown to be strongly influenced by the conceptual accessibility of the noun phrases in the utterance, including choices of active vs. passive voice, and double object vs. prepositional dative forms (e.g., Bock and Warren, 1985, Bock and Irwin, 1980, Bock and Loebell, 1990, Ferreira, 1994, McDonald et al., 1993, Bock, 1987). Two aspects of noun phrase accessibility are particularly relevant here because they can be directly linked to comprehension difficulty. One is noun animacy; English speakers have a tendency to locate animate concepts at initial positions in the sentence (Clark, 1965, Bates and MacWhinney, 1982, Bock, 1982, Bock, 1987, Bock et al., 1992). This observation has often been cast in terms of subjecthood because in English words that are mentioned first are strongly correlated with syntactic subjects (but see Prat-Sala and Branigan, 2000, De Smedt and Kempen, 1987, Kempen and Hoenkamp, 1987, Branigan et al., 2008). This tendency is particularly noticeable when speakers describe events with an animate and an inanimate participant in which the animate entity is the patient of the action, which results in an unusual preference for passive structures (The boy was hit by the truck is preferred over The truck hit the boy), even though the patient of an action is otherwise a natural candidate for the object position. Animate nouns are thus mapped onto subject syntactic positions in a way that appears independent of the agent or patient role of the nouns within the sentence (McDonald et al., 1993). This arrangement maximizes production incrementality: The earliest-planned (most accessible) element of the sentence is uttered first, while the speaker plans the less accessible upcoming material (De Smedt and Kempen, 1987, Kempen and Hoenkamp, 1987, Levelt, 1989).

Another aspect of noun phrases that modulates their relative accessibility in production is the thematic roles that nouns bear in the event referred to by the verb. Ferreira (1994) argued that the relative accessibility of the verb’s thematic roles influences the rate of active vs. passive sentence production. In production studies in which participants were given two nouns and a verb to make up a sentence, participants were more likely to passivize theme–experiencer verbs than regular agent–theme or experiencer-theme verbs when the nouns had the same or mismatching animacy. Theme-experiencer verbs such as surprise, please, or annoy denote an event in which something or someone causes a change of psychological state in the human animate participant – the experiencer (Belletti and Rizzi, 1988, Grimshaw, 1990, Levin and Rappaport, 1986, Cupples, 2002). Ferreira argued that the affected experiencer role assigned to a noun (e.g., to the mother in The child/gift pleased the mother) is more prominent than the theme-cause role (e.g., the child or gift), and that passive constructions result from the speakers’ preference to locate the most conceptually prominent noun in subject position (e.g., The mother was pleased by the child/gift). This preference contrasts with that of ordinary agent–theme verbs in which the agent invariably takes the sentential subject position, thus resulting in an active construction.

Production preferences in the domain of relative clauses have not been extensively investigated (but see Gennari, Mirkovic, & MacDonald, 2005), and little is known about the role of accessibility considerations in the production of these more complex structures. Relative clauses are thought to function like predicates or modifiers of a head noun, e.g., in the book that I bought, the relative clause that I bought modifies the noun book (Chierchia & McConnell-Ginet, 2000). According to some descriptions based on corpus studies, the discourse function of relative clauses is either to ground the head entity with respect to given information in the discourse or to provide a characterization of it (Fox & Thompson, 1990). Unlike main clauses in which passives and actives have different noun orders, the head of the relative clause invariably takes the initial position in the structure by virtue of discourse considerations and language-specific constraints (e.g., English is a head-first language). If the relative clause continues as an active object relative (e.g., the director that the movie pleased), the order of nouns (e.g. director, movie) is the same as in a passive relative (the director that was pleased by the movie). These contrasts suggest that some mechanism(s) different from those in main clauses may play a role in relative clause production: because the position of the head noun is fixed, accessibility may (as in main clauses) or may not influence different syntactic function assignment (e.g. subject vs. object) in relative clauses. Thus one goal of the production studies is to investigate the extent to which the accessibility factors that shape main clause structures also act in relative clauses. We address this question in the first section, which examines relative clause production using a method similar to the one used by Ferreira (1994) with main clauses. In Section 2, we use language corpora to ask whether the patterns identified in the production studies are representative of broad distributional regularities in the language. Finally, in Section 3, we ask about the effects of these distributional regularities on the comprehension of relative clauses.

Section snippets

Study 1: verb type in active vs. passive production preferences

In this study, we investigated the production preferences associated with theme–experiencer and agent–theme verbs when speakers are constrained to produce a relative clause with one animate and one inanimate noun. Because relative clauses and main clauses differ in several ways, the purpose of this study was to establish whether theme–experiencer verbs show a preference for passives when they occur within relative clauses, as Ferreira (1994) showed for these verbs in main clauses. To this end,

Study 3: verb type in active vs. passive production preferences

Previous corpus analyses of active vs. passive structures (Rohde, 2002) or of relative clause productions (Fox and Thompson, 1990, Roland et al., 2007) have not investigated the effect of verb type on structure within relative clauses. In this study we therefore investigated the tendencies of agent–theme and theme–experiencer verbs to occur in passive constructions within the context of relative clauses in several English corpora. We also investigated the properties of the specific verbs used

Study 5: verb types’ passive vs. active preferences

In our previous investigation of relative clause comprehension, we found that a key source of comprehension difficulty is ambiguity in how to assign thematic roles to the noun phrases that have been encountered (Gennari & MacDonald, 2008). In particular, the thematic roles that participants assigned to the nouns as the relative clause unfolds (as determined by gated completion studies) predicted comprehension difficulty (as indexed by reading times) at subsequent word positions. Infrequent and

General discussion

In several studies, we have shown how production preferences within object relative clauses are modulated by verb type and noun animacy, how production results in the laboratory are mirrored in broad language corpora, and how the production preferences were linked to comprehension difficulty. In Section 1, Study 1 found that theme–experiencer verbs are more likely to be produced in passive constructions than agent–theme verbs. In Study 2, the choice of passive structures varied with both verb

References (101)

  • D. Grodner et al.

    Syntactic complexity in ambiguity resolution

    Journal of Memory and Language

    (2002)
  • A. Kawamoto

    Nonlinear dynamics in the resolution of lexical ambiguity: A parallel distributed processing account

    Journal of Memory and Language

    (1993)
  • G. Kempen et al.

    An incremental procedural grammar for sentence formulation

    Cognitive Science: A Multidisciplinary Journal

    (1987)
  • J. King et al.

    Individual differences in syntactic processing: The role of working memory

    Journal of Memory and Language

    (1991)
  • R. Levy

    Expectation-based syntactic comprehension

    Cognition

    (2008)
  • R.L. Lewis et al.

    Computational principles of working memory in sentence comprehension

    Trends in Cognitive Sciences

    (2006)
  • M.C. MacDonald et al.

    Constraint satisfaction accounts of lexical and sentence comprehension

  • B. MacWhinney et al.

    The processing of restrictive relative clauses in Hungarian

    Cognition

    (1988)
  • W.M. Mak et al.

    The influence of animacy on relative clause processing

    Journal of Memory and Language

    (2002)
  • W.M. Mak et al.

    Animacy in processing relative clauses: The hikers that rocks crush

    Journal of Memory and Language

    (2006)
  • J.L. McDonald et al.

    Word and world order: Semantics, phonological, and metrical determinants of serial position

    Cognitive Psychology

    (1993)
  • E. McKone et al.

    Can generic expertise explain special processing for faces?

    Trends in Cognitive Sciences

    (2007)
  • K. McRae et al.

    Modeling the influence of thematic fit (and other constraints) in on-line sentence comprehension

    Journal of Memory and Language

    (1998)
  • M.J. Pickering et al.

    Do people use language production to make predictions during comprehension?

    Trends in Cognitive Sciences

    (2007)
  • M. Prat-Sala et al.

    Discourse constraints on syntactic processing in language production: A cross-linguistic study in English and Spanish

    Journal of Memory and Language

    (2000)
  • F. Reali et al.

    Processing of relative clauses is made easier by frequency of occurrence

    Journal of Memory and Language

    (2007)
  • D. Roland et al.

    Frequency of basic English grammatical structures: A corpus analysis

    Journal of Memory and Language

    (2007)
  • A. Sheldon

    The role of parallel function in the acquisition of relative clauses in English

    Journal of Verbal Learning and Verbal Behavior

    (1974)
  • M.F. St.John et al.

    Learning and applying contextual constraints in sentence comprehension

    Artificial Intelligence

    (1990)
  • L.M. Stallings et al.

    Phrasal ordering constraints in sentence production: Phrase length and verb disposition in heavy-NP shift

    Journal of Memory and Language

    (1998)
  • W. Tabor et al.

    Dynamical models of sentence processing

    Cognitive Science

    (1999)
  • M.K. Tanenhaus et al.

    Sentence comprehension

  • M.J. Traxler et al.

    Processing subject and object relative clauses: Evidence from eye movements

    Journal of Memory and Language

    (2002)
  • J.A. Van Dyke et al.

    Distinguishing effects of structure and decay on attachment and repair: A cue-based parsing account of recovery from misanalyzed ambiguities

    Journal of Memory and Language

    (2003)
  • J.B. Wells et al.

    Experience and sentence processing: Statistical learning and relative clause comprehension

    Cognitive Psychology

    (2009)
  • E. Wonnacott et al.

    Acquiring and processing verb argument structure: Distributional learning in a miniature language

    Cognitive Psychology

    (2008)
  • E. Bates et al.

    From first words to grammar: Individual differences and dissociable mechanisms

    (1988)
  • E. Bates et al.

    Functionalist approaches to grammar

  • A. Belletti et al.

    Psych-verbs and θ-theory

    Natural Language and Linguistic Theory

    (1988)
  • J.K. Bock

    Toward a cognitive psychology of syntax: Information processing contributions to sentence formulation

    Psychological Review

    (1982)
  • J.K. Bock

    Meaning, sound, and syntax: Lexical priming in sentence production

    Journal of Experimental Psychology: Learning, Memory and Cognition

    (1986)
  • J.K. Bock

    An effect of the accessibility of word forms on sentence structure

    Journal of Memory and Language

    (1987)
  • J.K. Bock et al.

    Syntactic effects of information availability in sentence production

    Journal of Verbal Learning and Verbal Behavior

    (1980)
  • J.K. Bock et al.

    Language production: Grammatical encoding

  • J.K. Bock et al.

    Framing sentences

    Cognition

    (1990)
  • J.K. Bock et al.

    From conceptual roles to structural relations: Bridging the syntactic cleft

    Psychological Review

    (1992)
  • J.K. Bock et al.

    Conceptual accessibility and syntactic structure in sentence formulation

    Cognition

    (1985)
  • D. Caplan et al.

    Verbal working memory and sentence comprehension

    Behavioral and Brain Sciences

    (1999)
  • F. Chang et al.

    Becoming syntactic

    Psychological Review

    (2006)
  • G. Chierchia et al.

    Meaning and grammar: An introduction to semantics

    (2000)
  • Cited by (0)

    View full text