Individual Differences in Syntactic Processing: The Role of Working Memory

Mecánica Teórica

23 pages
3 views

Please download to get full document.

View again

of 23
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Share
Description
Individual Differences in Syntactic Processing: The Role of Working Memory
Transcript
  JOURNAL OF MEMORY AND LANGUAGE 30,580-602 (1991) Individual Differences in Syntactic Processing: The Role of Working Memory JONATHAN KING AND MARCELADAMJUST Carnegie Mellon University The results of two experiments indicate that individual differences in syntactic processing are governed in part by the amount of working memory capacity available for language comprehension processes. Reading the verbs of an object relative sentence, such as The reporter that the senator attacked admitted the error, takes more time for readers with less working memory capacity for language, and their resulting comprehension is less accurate. Experiment 1 investigated the effects of a concurrent working memory load and found that with no load or a small memory load many Low Span readers comprehended object relative sentences very poorly although their reading times in the critical area of these sentences were greater than those of High Span subjects. Experiment 2 replicated the reading time effects of Experiment 1 for object relative sentences and showed that pragmatic information improved the comprehension of the lower capacity readers, although their use of this in- formation was limited to the clause in which it was presented. o IW Academic press, IIIC. In this paper, we develop the hypothesis that individual differences in syntactic pro- cessing can be observed in normal readers, and that these differences are influenced by differences in working memory capacity for language. Although the processing of syn- tax has been an active topic of language research for several decades, only a few re- searchers have examined individual differ- ences in the processing of syntactic infor- mation (e.g., Gleitman & Gleitman, 1970; Holmes, 1987). The present research inves- tigates how syntactic processing and the re- sulting comprehension of sentences may be influenced by a reader’s working memory capacity for language. Working memory ca- pacity includes not only storage functions for intermediate or partial products, but a flexibly deployable pool of operational re- sources needed to perform the symbolic computations that generate those interme- This work was supported in part by Grant MH- 29617 from NIMH, and Research Scientist Develop- ment Award MH-00662 from NIMH. We thank Patri- cia Carpenter and Brian MacWhinney for comments on the manuscript. Requests for reprints should be addressed to Marcel Adam Just, Psychology Depart- ment, Carnegie Mellon University, Pittsburgh, PA 15213. diate and partial products (Daneman & Car- penter, 1980). Individual differences in working memory capacity may be the result of differences in either the size of the pool, the efficiency of the processes that perform symbolic computations, or both. There are several reasons for hypothesiz- ing a relation between syntactic processing and working memory. Syntactic processing transforms a linear sequence of words into a nonlinear (hierarchical) syntactic struc- ture, and this transformation requires the temporary storage of word representations during the left-to-right processing of a sen- tence. Moreover, linguistic parsing models have appealed to working memory storage limitations to account for the preferences among alternative interpretations in tempo- rarily ambiguous sentences (e.g., Kimball, 1973; Frazier 8z Foder, 1978; Gibson, 1990). Limitations on the processing func- tions of working memory are similarly im- plicated in syntactic processing by evi- dence that certain syntactic gaps are diffi- cult to find and process (e.g., MacDonald, 1989). The classic example of a syntactic struc- ture that makes large demands on working memory capacity is a center-embedded rel- 580 0749-5%x/91 $3.00 Copy&ht 0 1991 by Academic Press, Inc. AU rights of reproduction in any form reserved.  DIFFERENCES IN SYNTACTIC PROCESSING 581 ative clause, such as the one in sentence (1) below. It is also called an object relative for brevity, a name that reflects the role that the head noun plays as the object in the relative clause. (1) The reporter that the senator at- tacked admitted the error. Subjects who hear a sentence like (1) and then try to paraphrase it make errors in matching verbs with their agents approxi- mately 15% of the time (Larkin & Burns, 1977). There are three kinds of demands such sentences make on working memory that combine to make them difficult to com- prehend. First, the embedded clause in such sentences interrupts the main clause, drawing on the storage resources of work- ing memory. The representation of the clause segment that precedes the interrup- tion must either be retained in working memory during the processing of the em- bedded clause or be reactivated at the con- clusion of the embedded clause (Miller & Chomsky, 1963; Wanner & Maratsos, 1978). Second, the assigning of the proper thematic roles to the two noun phrases im- poses some difficulty. In particular, know- ing whether the head of the relative clause is the agent or the patient of the relative clause verb causes comprehension of the relative clause to be less accurate than comprehension of the main clause in this kind of sentence (e.g., Holmes & O’Regan, 1981). Third, the assignment of two differ- ent roles to a single syntactic constituent also taxes working memory capacity. In the example above, reporter is the agent in one clause and the recipient of the action in the other. Associating a single concept with two different roles simultaneously seems to be a source of difficulty in language com- prehension (Bever, 1970) and the switch- ing of perspective in the construction of such a concept can also tax cognitive re- sources (MacWhinney & Pleh, 1988). The difficulty of processing an object rel- ative sentence contrasts with the relative ease of processing a related construction, called a subject relative. In a subject rela- tive sentence, like (2) below, the main clause is interrupted, but role assignments can be made one at a time, and the constit- uents have parallel roles in the two clauses. (2) The reporter that attacked the sena- tor admitted the error. These subject relative sentences are both easier to comprehend and easier to process on-line than object relative sentences like (1) (Holmes & O’Regan, 1981). In the read- ing of the easier subject relative sentences like (2), the head of the relative clause needs to be maintained over only a short distance, since the agent role can be as- signed to the reporter as soon as the verb attacked is processed; this leaves only a single role to assign when the senator is reached. This contrasts with the reading of an object relative sentence like (1) where the reader cannot assign a thematic role to either the reporter or the senator until the verb attacked is encountered. One would also expect a subject relative sentence like (2) to be easier, because the agent in the relative clause is also the agent in the main clause, and thus requires no assignment of conflicting roles to the actor and no per- spective shift. Thus, the three aspects that make the processing of an object relative sentence demanding of working memory are all mitigated in a subject relative sen- tence. Our main hypothesis is that the dif- ference between these two sentence types in their working memory requirements will interact with the differences in working memory capacity we can observe in a col- lege level population. The relation between working memory capacity and language processing has been addressed by the work of Daneman and Carpenter (1980, 1983) at the referential level of processing, but not at the syntactic level. Daneman and Carpenter have shown that differences in a measure of working memory capacity predict substantial differ- ences in the ability to compute the referents of pronouns and the ability to integrate se- mantic information between and within sentences. The measure of working mem-  582 KING AND JUST ory capacity developed by Daneman and Carpenter, called the Reading Span Test, determines the longest set of sentences (of approximately 15 words each) that subjects can orally read and from which they can then recall all the sentence-final words of the set. This task requires subjects to main- tain the set of unrelated sentence-final words from preceding sentences while they process each successive sentence in the set. Performance on the Reading Span Test accounts for almost 35% of the variance on verbal SAT scores and 80% of the variance in performance on a task which measures the ability to associate pronouns with their antecedents over varying distances (Dane- man & Carpenter, 1980; see also Turner & Engle, 1989). By contrast, simple word or digit span tests, which make little demand on any operational resources related to comprehension, typically are not substan- tially correlated with reading ability or other interesting language tasks (Perfetti 8z Lesgold, 1977). The advantage of the Read- ing Span measure over the simpler mea- sures may derive from a similarity between the reading span task and more interesting linguistic tasks in their demands on the storage and processing functions of work- ing memory. Our main hypotheses are that while all readers should have greater diffi- culty with object relative sentences than with subject relative sentences, it is the readers with less working memory capacity for language as assessed by the Reading Span test who should have the most difli- culty. The distribution of processing load dur- ing comprehension should be reflected in the word-by-word reading times for object relative and subject relative sentences like (1) and (2): (1) The reporter that the senator at- tacked admitted the error. (2) The reporter that attacked the sena- tor admitted the error. We assume that readers try to interpret each word as soon as they encounter it, in agreement with the Immediacy of Interpre- tation Hypothesis (Just & Carpenter, 1980, 1987). In the case of the object relative sen- tence such as (1) the increased processing demands first manifest themselves when the reader reaches the first verb (attacked). At that point, the head noun (reporter) must be assigned to the thematic role of patient and senator must be assigned to the the- matic role of agent. Next, the reader en- counters the verb (admitted) and must find an agent for this verb, but this agent is not the same as the agent of attacked. Thus the two adjacent verbs should be the locus of the extra processing load. This result was found by Ford (1983) with a continuous lex- ical decision task in which readers judged each successive word of a sentence to de- termine whether it was a real word or a nonword. Subjects took 25 ms longer to make a lexical decision on the verbs at- tacked and admitted in an object relative sentence like (1) than in a subject relative sentence like (2). Similar results were ob- tained by Holmes and O’Regan (1981) in a reading task which measured the duration of subjects’ eye fixations using sentences in French, which has a similar subject/object relative clause distinction. Consequently, it is reasonable to expect that word-by-word reading times will be sensitive to syntactic processing load and to working memory ca- pacity differences among individuals. It is also known that the process of role assign- ment in object relative sentences is difficult enough to cause a substantial number of comprehension errors, which suggests that comprehension scores could also be sensi- tive to working memory capacity differ- ences. The current studies also experimentally manipulated the subjects’ effective working memory capacity by either imposing an ex- traneous memory load during the process- ing of the sentences (Experiment l), or by supplying pragmatic information that would make it easier to comprehend object rela- tive sentences (Experiment 2). In previous studies that used a memory load manipula- tion, Baddeley (1986) examined sentence comprehension that was concurrent with  DIFFERENCES IN SYNTACTIC PROCESSING 583 retaining a string of digits or articulating nonsense syllables. The concurrent storage task imposes an additional load which con- sumes some of the resources of working memory; consequently, it is expected to de- grade some facet of performance in a de- manding comprehension task. By contrast, when additional pragmatic information is supplied, comprehension performance should improve, because the pragmatic in- formation could obviate some of the syn- tactic processing usually required by object relative sentences. EXPERIMENT 1 This experiment investigated how differ- ences in the working memory capacity of readers affected the processing of syntactic structure. The experimental task itself var- ied two factors: the syntactic structure of the sentences read and whether or not a memory load was carried in working mem- ory concurrently while reading. The third and central factor studied was the working memory capacity of the subjects, which was assessed by the Reading Span task. The structure of the reading task was simi- lar to that of the Reading Span task de- scribed in the introduction. Subjects read sets of one, two, or three sentences and were asked to recall the sentence final words at the end of the set. The syntactic factor was manipulated by varying the structure of the final, target sentence. Half of these targets included subject relative clauses, while the other half contained more difficult object relative clauses. The external memory load factor was manipu- lated by varying the number of sentences preceding the target. In addition, subjects were required to respond to a true-false probe item testing their comprehension of the final target sentence in the set. This de- sign allowed us to measure the proportion of sentence-final words recalled, the pro- portion of comprehension probes answered correctly, and the reading times on each word of the target sentence. While recall of sentence-final words will primarily be a re- liability check on our use of the Reading Span Test, sentence comprehension accu- racy and reading time results will be crucial to our hypothesis. This is particularly true for one- and two-sentence trials, which should be within the capacity limitations of all subjects. Our hypothesis is that when the combi- nation of storage and processing require- ments of a comprehension task exceeds working memory capacity, performance on one or more aspects of the task will deteri- orate. Thus, when comparing performance between groups of subjects differing in working memory capacity for language, we would expect the lower capacity subjects to show lower comprehension accuracy for syntactically more complex object relative sentences and increased reading times at the sentence locations in object relative sentences where substantial syntactic com- putations must be performed. This pattern of results can be expected from the in- creased processing demands of the object relative sentences compared to subject rel- ative sentences, as described above. The effect of a working memory load should de- pend on the working memory capacity of the subjects themselves. The performance of high capacity subjects maintaining a working memory load should more closely resemble the performance of low capacity subjects without any load. Similarly, the performance of low capacity subjects main- taining a working memory load should be- come worse up until the point when their capacity is entirely exhausted, at which point patterns of performance may cease to be easily interpretable. Method Materials. Each target sentence occurred as the final sentence of a set of either 1, 2, or 3 sentences. In a set with one sentence, the subject read only the target sentence; then he or she recalled the last word of the sentence and then answered a question about the sentence. In a set of two sen- tences, there was one sentence prior to the  584 KING AND JUST target sentence. The subject read each of the two sentences, recalled the two sen- tence-final words, and then answered a question about the target sentence. Three sentence trials were the same, except that all three sentence-final words were re- called. For convenience, we will refer to these conditions as sets of 1, 2, or 3 words. However, it is important to remember that when the subject was reading the target sentence, she or he was retaining only 0, 1, or 2 words, respectively, from the preced- ing sentences. Thus, the number of words retained during the reading of the critical sentence was one less than the number of sentences in the set. There were 60 trials in the experiment, evenly divided among sets consisting of 1, 2, or 3 sentences. There were 36 experi- mental sets and 24 tiller sets. Half of the experimental sets contained an object rela- tive sentence as the last sentence in the trial, while the other half contained a sub- ject relative sentence as the last sentence. The experimental and filler trials were pre- sented in a different random order for each subject. All target sentences were con- structed so that the relative clauses were completely reversible and both the correct and reversed versions were equally plausi- ble and appropriate in the context of the sentence. All the sentences were between 12 and 17 words in length. An example of an object relative sentence is: The reporter that the senator attacked admitted the error publicly after the hear- ing. Its subject relative counterpart would be: The reporter that attacked the senator admitted the error publicly after the hear- ing. Eighteen familiar transitive verbs that take animate subjects and objects were each used twice (with different grammatical sub- jects and objects) in the embedded clauses of the target sentences. Each of the sen- tences ended with an extra prepositional or adverbial phrase that followed the direct object of the main verb. This was done so that the reading time on the direct object was not contaminated by any effect due to encountering the end of the sentence, where subjects sometimes pause briefly (Just & Carpenter, 1980). The filler sen- tences that preceded the last sentence in either the experimental or filler sets were sentences that had previously been found to be fairly easy to process, as measured by per-character reading times. The final sen- tence of each filler set was chosen from an- other set of sentences which had previously been found to be rather difficult, in order to reduce the likelihood of subjects noticing the object relative sentences as being more difficult than the tiller trials. An example of this type of sentence is: To equivocate on issues that are mal- adaptive for mankind is a futile objective. The comprehension test probe for a tar- get sentence was constructed by combining one of the two verbs in the sentence with two of the three nouns. All pragmatically possible orders were used in equal numbers to defeat guessing strategies. Possible com- prehension probes for the object relative sentence (1) would include any of the fol- lowing sentences: (1) The reporter that the senator at- tacked admitted the error . . . The reporter attacked the senator. (False) The senator attacked the reporter. W-N The reporter admitted the error. (True) The senator admitted the error. (False) The same four statements served as the comprehension test items for the subject relative sentence, except that the two state- ments probing the relative clause would have opposite truth values. The number of true and false comprehension test items was balanced across clause type and mem-
Advertisement
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks