David J. Townsend

Sentence Comprehension: The Integration of Habits and Rules

Using sentence comprehension as a case study for all of cognitive science, David Townsend and Thomas Bever offer an integration of two major approaches, the symbolic-computational and the associative-connectionist. The symbolic-computational approach emphasizes the formal manipulation of symbols that underlies creative aspects of language behavior. The associative-connectionist approach captures the intuition that most behaviors consist of accumulated habits. The authors argue that the sentence is the natural level at which associative and symbolic information merge during comprehension. The authors develop and support an analysis-by-synthesis model that integrates associative and symbolic information in sentence comprehension. This integration resolves problems each approach faces when considered independently. The authors review classic and contemporary symbolic and associative theories of sentence comprehension, and show how recent developments in syntactic theory fit well with the integrated analysis-by-synthesis model. They offer analytic, experimental, and neurological evidence for their model and discuss its implications for broader issues in cognitive science, including the logical necessity of an integration of symbolic and connectionist approaches in the field.

Interclause relations and clausal processing

In two experiments subjects were interrupted while listening to a two-clause sentence just before the last word of either the initial clause or the final clause. In Experiment I subjects were timed on their decision about whether a verb—object phrase was consistent in meaning with the sentence fragment they had just heard. Overall these decisions were made more quickly when a main clause was interrupted than when a subordinate clause was interrupted, but the size and direction of main—subordinate differences varied with the causal—temporal properties of subordinate clauses. In Experiment II subjects were timed on their decisions about whether a particular probe word had occurred in the sentence fragment. Target position effects differed for main and subordinate clauses, but again, these effects were related to causal—temporal relations between clauses. The two experiments together suggest that interclause semantic relations affect the immediate processing of clauses.

Thematic processing in sentences and texts

Abstract Since many of the semantic relations that exist between clauses in complex sentences also exist between sentences in texts, a description of complex sentence processing should generalize to sentence processing in texts. This paper explores the usefulness of such a generalization for describing the online processing of sentences in narratives. Four experiments on the processing and retention of isolated complex sentences showed that connectives that signal a disruption in the causal and temporal order of propositions influence on-line processing and recall. Four additional experiments on the processing and retention of sentences in narratives showed that causal/ temporal relations within complex sentences and between sentences affect sentence reading time, and the immediate and long term memory organization of propositions. The results sugge partially independent processing systems for both isolated complex sentences and sentences in texts. One set of processes operates on superficial information to obtain a literal propositional representation; another operates on propositional information to obtain a thematic representation of the sentence or text. In general, factors that obscure the thematic relevance of a clause or sentence also indirectly affect the processing of propositional meaning.

Familial handedness and access to words, meaning, and syntax during sentence comprehension

We compared right-handed familial dextral (FS-) and familial sinistral (FS+) participants who were aged either 10-13 years (children) or 18-23 years (adults). In word probe and associative probe tasks, FS+ adults responded faster than all other groups and FS+ children responded more slowly than all other groups. In the word probe task, only the FS- adults showed a significant effect of the serial position of the target word. We interpret these differences to support an analysis-by-synthesis model of comprehension in which individuals who differ in familial handedness and age emphasize different linguistic representations during comprehension. In general, FS+ individuals focus on words and meaning, while FS- individuals focus on syntactic representations. In FS+ individuals, age-related experiences with language produce a shift in responding from compositional meaning to words and their associations. In FS- individuals, age-related experiences with language produce a shift toward responding based more on detailed syntactic representations, including the serial order of words and possibly the structural roles of clauses.

THE EMPEROR'S PSYCHOLINGUISTICS

We discuss and debunk five common assumptions about the interrelation of semantics, syntax, and frequency during sentence processing. In the course of this, we explore the implications of the view that syntax is assigned as the last stage of comprehension rather than the first: Statistically based perceptual strategies propose an initial semantic representation, which then constrains the assignment of syntactic representations. This view accounts for a variety of facts, as well as suggesting some surprising new ones.

Natural units of representation interact during sentence comprehension

Subjects read aloud the verb IS and ARE that was projected to interrupt a spoken sentence: immediately preceding the projected word, the sentence contianed a “verb” + ing noun” sequence, which was either a gerund (“raking leaves”), an adjectival phrase (“diving submarines”) or ambiguous (“growing flowers”). The experiment investigated the effect on word-reading time of various independent variables in the preceding context clause that biased the subject to expect either a gerund or adjectival construction. The operational measure of the effectiveness of a context was the difference in reading time between a word consistent with the bias and a word inconsistent with the bias. The stronger effects occurred when the “verb + ing noun” phrase was ambiguous and (1) the plural/singular dimension in the preceding context clause was morphologically explicit, (2) the context clause was introduced by THOUGH (as opposed to IF, or being a main clause), (3) there was an explicit form of BE in the context clause, (4) the biased interpretation from the context clause was the gerund. These effects and others support a structurally differentiated model of comprehension in which listeners link representations at distinct levels of representation expressed in the units that are most natural at that level (e.g., propositions at the semantic level, words and word-groupings at the syntactic level).

The development of complex sentence processing strategies

Abstract Children aged 3, 4, and 5 years and adults heard sentences with clauses connected by after, and, or before, saw a picture, and indicated whether or not the picture matched one of the events of the sentence. Response times were taken as a measure of immediate accessibility to the meaning of the clause that the picture was about. Temporal organization of sentence meanings was dominant in 3-year-olds and adults, but not in 4- or 5-year-olds. The 3-year-olds and especially the adults processed and-sentences as implicitly temporal. The results for 4- and 5-year-olds are interpreted as indicating experimentation with alternate strategies for organizing sentences based on the structural/presuppositional properties of clauses.

Aspectual Coercion in Eye Movements

Comprehension includes interpreting sentences in terms of aspectual categories such as processes (Harry climbed) and culminations (Harry reached the top). Adding a verbal modifier such as for many years to a culmination coerces its interpretation from one to many culminations. Previous studies have found that coercion increases lexical decision and meaning judgment time, but not eye fixation time. This study recorded eye movements as participants read sentences in which a coercive adverb increased the interpretation of multiple events. Adverbs appeared at the end of a clause and line; the post-adverb region appeared at the beginning of the next line; follow-up questions occasionally asked about aspectual meaning; and clause type varied systematically. Coercive adverbs increased eye fixation time in the post-adverb region and in the adverb and post-adverb regions combined. Factors that influence the appearance of aspectual coercion may include world knowledge, follow-up questions, and the location and ambiguity of adverbs.

The use of higher-level constraints in monitoring for a change in speaker demonstrates functionally distinct levels of representation in discourse comprehension

Abstract We test and disprove the common assumption that pragmatic probability facilitates the processing of lower linguistic levels. Two experiments show that detection of the acoustic properties that distinguish two speakers is harder in sentences that are pragmatically more probable. At the same time. detection of the same acoustic properties is easier at later points in a clause than at earlier points. The pragmatic inhibition in detecting acoustic properties suggests that different levels of linguistic processing compete for processing resources, while the within-sentence facilitation suggests limited interactions between processes at adjacent levels. The two results together demonstrate that discourse-level and sentence-level representations are functionally distinct during comprehension, and they force a modification of the view that linguistic processes at different levels of representation occur in architecturally segregated “modules” that do not share resources. We argue that two levels of repre...

Word-monitoring tasks interact with levels of representation during speech comprehension

Researchers frequently use data from monitoring tasks to argue that constraints on meaning facilitate lower-level processes. An alternate hypothesis is that the processing level that a monitoring task requires interacts with discourse-level processing. Subjects monitored spoken sentences for a synonym (semantic match), a nonsense word (phonological match), or a rhyme (phonologically and semantically constrained matching). The critical targets appeared at the beginning of the final clause in two-clause sentences that began with if, which signals a semantic analysis at the discourse level, or with though, which maintains a surface representation. Synonym-monitoring times were faster for if than for though, nonsense word-monitoring times were faster for though than for if, and rhyme-monitoring times did not differ for if and though. The results show that conjunctions influence how listeners allocate attention to semantic versus phonological information, implying that listeners form these kinds of information independently.