Patrick Sturt

Depth of processing in language comprehension: not noticing the evidence

The study of processes underlying the interpretation of language often produces evidence that they are complete and occur incrementally. However, computational linguistics has shown that interpretations are often effective even if they are underspecified. We present evidence that similar underspecified representations are used by humans during comprehension, drawing on a scattered and varied literature. We also show how linguistic properties of focus, subordination and focalization can control depth of processing, leading to underspecified representations. Modulation of degrees of specification might provide a way forward in the development of models of the processing underlying language understanding.

The time-course of the application of binding constraints in reference resolution

Abstract We report two experiments which examined the role of binding theory in on-line sentence processing. Participants’ eye movements were recorded while they read short texts which included anaphoric references with reflexive anaphors ( himself or herself ). In each of the experiments, two characters were introduced into the discourse before the anaphor, and only one of these characters was a grammatical antecedent for the anaphor in terms of binding theory. Both experiments showed that Principle A of the binding theory operates at the very earliest stages of processing; early eye-movement measures showed evidence of processing difficulty when the gender of the reflexive anaphor mismatched the stereotypical gender of the grammatical antecedent. However, the gender of the ungrammatical antecedent had no effect on early processing, although it affected processing during later stages in Experiment 1. An additional experiment showed that the gender of the ungrammatical antecedent also affected the likelihood of participants settling on an ungrammatical final interpretation. The results are interpreted in relation to the notions of bonding and resolution in reference processing.

Linguistic focus and good-enough representations: An application of the change-detection paradigm

A number of lines of study suggest that word meanings are not always fully exploited in comprehension. In two experiments, we used a text-change paradigm to study depth of semantic processing during reading. Participants were instructed to detect words that changed across two consecutive presentations of short texts. The results suggest that the full details of word meanings are not always incorporated into the interpretation and that the degree of semantic detail in the representation is a function of linguistic focus. The results provide evidence for the idea that representations are only good enough for the purpose at hand (Ferreira, Bailey, & Ferraro, 2002).

Processing coordinated structures: incrementality and connectedness

We recorded participants eye movements while they read sentences containing verb-phrase coordination. Results showed evidence of immediate processing disruption when a reflexive pronoun embedded in the conjoined verb phrase mismatched the sentence subject. We argue that this result is incompatible with models of human parsing that employ only bottom-up parsing procedures, even when flexible constituency is employed. Models need to incorporate a mechanism similar to the adjoining operation in Tree-Adjoining Grammar, in which one structure is inserted into another.

Linguistic focus and memory: An eye movement study

We report an eyetracking study investigating the effects of linguistic focus on eye movements and memory during two readings of a text. Across two presentations of the text, a critical word either changed to a semantically related word or remained unchanged. Focus on the critical word was manipulated using context. Eye movements were monitored during reading, and there was a secondary task of detecting the word change. Results indicated that when a word changed, participants were more successful at detecting it when it was in focus. In the second display, there were more fixations and longer viewing times on a changed than on an unchanged word, but only when the critical word was in focus; eye movement data for changed and unchanged words did not differ when the word was not in focus. We suggest that linguistic focus leads to more detailed lexical semantic representations but not more effortful initial encoding of information.

Learning first-pass structural attachment preferences with dynamic grammars and recursive neural networks

One of the central problems in the study of human language processing is ambiguity resolution: how do people resolve the extremely pervasive ambiguity of the language they encounter? One possible answer to this question is suggested by experience-based models, which claim that people typically resolve ambiguities in a way which has been successful in the past. In order to determine the course of action that has been successful in the past when faced with some ambiguity, it is necessary to generalize over past experience. In this paper, we will present a computational experience-based model, which learns to generalize over linguistic experience from exposure to syntactic structures in a corpus. The model is a hybrid system, which uses symbolic grammars to build and represent syntactic structures, and neural networks to rank these structures on the basis of its experience. We use a dynamic grammar, which provides a very tight correspondence between grammatical derivations and incremental processing, and recursive neural networks, which are able to deal with the complex hierarchical structures produced by the grammar. We demonstrate that the model reproduces a number of the structural preferences found in the experimental psycholinguistics literature, and also performs well on unrestricted text.

A new look at the syntax-discourse interface: the use of binding principles in sentence processing.

Within Generative Grammar, binding constraints on co-reference are usually defined in syntactic terms. However, some researchers have pointed out examples in which syntactically defined binding constraints do not seem to apply, proposing instead that a complete account of linguistic co-reference needs to consider notions of discourse structure. There have been several proposals in the literature for the division of labor between syntax and discourse in the definition of binding constraints. In this paper, we review these proposals in the context of recent work that applies on-line techniques to explore the roles of syntactic and discourse preferences in terms of the time course with which they become active during sentence comprehension. Some of this research suggests that (syntactic) binding principles may be momentarily applied during processing, even in cases in which the final interpretation suggests otherwise. We end the paper by considering the theoretical and methodological implications of this view.

Competence and performance grammar in incremental processing

The goal of this paper is to explore some consequences of the dichotomy between competence and performance from the point of view of incrementality. We introduce a TAG-based formalism that encodes a strong notion of incrementality directly into the operations of the formal system. A left-associative operation is used to build a lexicon of extended elementary trees. Extended elementary trees allow derivations in which a single fully connected structure is mantained through the course of a left-to-right word-by-word derivation. In the paper, we describe the consequences of this view for semantic interpretation, and we also evaluate some of the computational consequences of enlarging the lexicon in this way.

Se servir du mouvement des yeux durant la lecture comme mesure implicite de l'acceptabilité des extensions de marque

Nous utilisons les techniques de suivi du mouvement des yeux pour étudier le degré dacceptabilité des consommateurs envers différentes extensions de marque. Des phrases étaient présentées aux participants (par exemple: Jai voulu enregistrer une chanson sur Polaroid mais ça coûte trop cher), et ces phrases ne pouvaient être totalement comprises que si les répondants imaginaient lextension de marque sous-entendue. À partir des travaux sur le traitement des expressions métonymiques (Frisson et Pickering, 1999), nous suggérons que les extensions de marque plausibles poseront peu de difficultés lors de la première lecture, alors que les extensions invraisemblables perturberont immédiatement la lecture. Lanalyse des enregistrements des mouvements des yeux confirme ces prédictions. Les extensions plausibles ont engendré des difficultés mineures lors de la lecture de la fin de la phrase, ce que nous expliquons comme la construction dune nouvelle interprétation de la marque. En revanche, les extensions invraisemblables ont immédiatement généré des perturbations, ce que nous expliquons comme une difficulté à fournir une interprétation cohérente.

Incrementality in syntactic processing: computational models and experimental evidence: keynote talk

It is a well-known intuition that human sentence understanding works in an incremental fashion, with a seemingly constant update of the interpretation through the left-to-right processing of a string. Such intuitions are backed up by experimental evidence dating from at least as far back as Marslen-Wilson (1973), showing that under many circumstances, interpretations are indeed updated very quickly. n nFrom a parsing point of view it is interesting to consider the structure-building processes that might underlie incremental interpretation---what kinds of partial structures are built during sentence processing, and with what time-course? n nIn this talk I will give an overview of the state-of-the-art of experimental psycholinguistic research, paying particular attention to the time-course of structure-building. The discussion will focus on a new line of research (some as yet unpublished) in which syntactic phenomena such as binding relations (e.g., Sturt, 2003) and unbounded dependencies (e.g., Aoshima, Phillips, & Weinberg, in press) are exploited to make a very direct test of the availability of syntactic structure over time. n nThe experimental research will be viewed from the perspective of a space of computational models, which make different predictions about time-course of structure building. One dimension in this space is represented by the parsing algorithm used: For example, within the framework of Generalized Left Corner Parsing (Demers, 1977), algorithms can be characterized in terms of the point at which a context-free rule is recognized, in relation to the recognition-point of the symbols on its right-hand side. Another relevant dimension is represented by the type of grammar formalism that is assumed. For example, with bottom-up parsing algorithms, the degree to which structure-building is delayed in right-branching structures depends heavily on whether we employ a traditional phrase-structure formalism with rigid constituency, or a cateogorial formalism with flexible constituency (e.g., Steedman, 2000). n nI will argue that the evidence is incompatible with models which predict systematic delays in the construction of syntactic structure. In particular, I will argue against both head-driven strategies (e.g., Mulders, 2002), and purely bottom-up parsing strategies, even when flexible constituency is employed. Instead, I will argue that to capture the data in the most parsimonious way, we should turn our attention to those models in which a fully connected syntactic structure is maintained throughout the processing of a string.