Click location and syntactic structure

Four experiments are reported in which Ss had to judge the location of clicks superimposed on recorded sentences. The first experiment showed that the accuracy of locating the clicks was a function of the position of the click in the constituent structure, the greatest accuracy being for clicks at major clause boundaries. The second experiment showed that this effect was independent of migration, i.e., the tendency for judgments to be displaced towards the major clause break. In the third experiment, it was shown that the requirement that S reproduce the sentence did not influence the response distribution. Finally, in the fourth experiment, a small but significant trend for location accuracy to decrease with decreasing separation of the click from the major break was found. However, this trend was much smaller than the differences in accuracy for various positions in the constituent structure. It was concluded that click location accuracy can be used as an index of perceptual processing load during recognition of individual sentences.     

Detection of extraneous signals during sentence recognition

In two separate experiments, 40 Ss were presented with recorded sentences during each of which a click occurred. Ss had to depress a key as soon as they heard the click. RTs were f aster when the click was located at the major syntactic break of the sentence compared with RTs to clicks not at a break. This confirmed the hypothesis that processing load is a function of the surface structure of sentences, although the role of minor breaks was not clear. A second finding was that RTs were slower when the click was in the first rather than in the second half of the sentence. This can also be explained in terms of differential processing loads.     

On catching on to idiomatic expressions

These experiments concern the comprehension of idiomatic expressions. The hypothesis was that there are distinct idiomatic and literal modes of processing sentences. In two experiments, 414 undergraduates read a series of sentences containing either literal or idiomatic ambiguities and then a test which had both a literal and an idiomatic meaning. Ss recorded, which meaning they perceived first. Taken together, the experiments indicate that inducing a set to perceive idioms can increase the proportion of people seeing the idiomatic meaning of test sentence first and a set to perceive literal meanings can reduce this proportion compared to a no-set baseline. Since the procedures to induce set did not involve grammatical or semantic information relevant to comprehension of the test sentence, these results suggest the existence of distinct literal and idiomatic processing strategies.     

The comprehension and verification of ambiguous sentences

When Ss are presented with an ambiguous sentence they tend to interpret it in only one way. If later events warrant, Ss can recover the other meaning, a process which takes time. These conclusions follow from the results of a study in which 40 undergraduate Ss verified whether or not pictures shown at the end of a sentence represented the meaning of the sentence. When ambiguous sentences were presented, the verification time (VT) was no slower than for unambiguous sentences if the picture represented the “expected” meaning (as determined on a pre-test) of the ambiguity. The VT to the picture representing the “unexpected” meaning of the ambiguity was longer than VT to corresponding control sentences.     

Click monitoring revisited: An on-line study of sentence comprehension

Spoken sentence comprehension is based upon rapid and complex psychological processes, yielding a constantly fluctuating cognitive load. The aim of this study was to evaluate on-line click monitoring, a classical but poorly exploited experimental method, which should allow for an easy measurement of processing load at any chosen point of experimental sentences. In Experiments 1 and 2, we obtained longer latencies to clicks located at the boundary of reversible object relative clauses than to clicks identically located in subject relatives and to clicks located earlier within object relatives. Experiment 3 further revealed that this effect of syntactic type was specific to transposed object relatives and did not occur with normal object relatives. In Experiment 4, we observed longer latencies with semantically reversible than with irreversible sentences, but no difference between actives and passives. These results were obtained under strict control of potential lexical and phonological biases, and suggest that on-line click monitoring may be one useful tool in the study of sentence comprehension.     

Effects of task on the perceptual organization of sentences

Children and college students listened to sentences and responded to a question or a probe word after each sentence. To prevent Ss from using different perceptual strategies in performing the two tasks, question items and probe items were presented in mixed lists. In the probe task, the relationship between sentence structure and correct response latencies resembled the findings of an earlier study with block lists. The results from the question task were less similar to those obtained earlier, perhaps because, with mixed lists, Ss tended to organize all sentences in a manner especially suited to the probe task. A different type of organization might be used in the process of comprehending a sentence solely for the purpose of answering questions.     

Generation of a pictorial code

In Experiments I and II direct measures were obtained of the time required to convert a picture or a sentence to a report of visualization or to a simple drawing, latencies were faster for pictures than for sentences and were affected by the surface form of the sentence. In Experiment Ill, Ss matched pictures or sentences against a test picture under conditions of simultaneous or successive presentation. “Yes” and “No” response times were affected by sentence form under both conditions, although the difference between sentence-picture and picture-picture comparisons was virtually eliminated under the successive condition. In Experiment IV, Ss held a sentence or picture in memory over a 1,500-msec interval at the end of which a test picture or an instruction to draw the design was presented. Construction latencies were shorter in response to pictures than sentences, and there were effects of sentence form on verification time. Some implications of these results are discussed.     

Processing strategies in picture-sentence verification tasks

The encoding and comparison strategies used by observers in matching pictures and sentences were investigated. The location of the mismatch or difference between a sentence and a picture was varied, and the latencies of "same"-"different" responses were compared as a function of whether the mismatch occurred in the sentence subject, verb, or object. Sentences were presented auditorily and varied in both voice and reversibility. Pictures either preceded or were presented simultaneously with the sentences. In all cases, Ss adopted a serial self-terminating comparison strategy and a surface structure encoding strategy: The comparison was terminated as soon as a mismatch was encountered, and actives were processed in the order subject-verb-object, whereas passives were processed in the order object-verb-subject.     

Positive forgetting of sentence material

This experiment investigated the positive-forgetting phenomenon with sentence material. Sets of sentences were presented to Ss with each sentence being cued “remember” or “forget” immediately following its presentation. To-be-remembered (TBR) sentences were found to be more accessible thanto-be-forgotten (TBF) sentences and uncued control sentences. Sentence connectedness was found to be an important determiner of the magnitude of the observed recall phenomenon; but differential sentence interest was not a significant factor. Using a multiple-choice recognition test, key words or phrases from the to-be-forgotten sentences were found to be equally available as key words or phrases from the to-be-remembered sentences. These results warrant an extention of Bjork’s (1970, 1972) selective-rehearsal and differential-grouping interpretation of the positive-forgetting phenomenon to encompass sentence material. Some implications of the findings for single-presentation information acquisition were noted.     

The two meanings of a homophone

Massachusetts Institute of Technology, Cambridge, Massachusetts 02139 Ss were presented with sentences whose main noun was either a homophone or not. Following each sentence, Ss were presented with a single noun and they had to indicate whether it was relevant to the sentence or not. More specifically, irrelevant probes following sentences containing homophones were either irrelevant with respect to both interpretations of the homophone or irrelevant with respect to the present interpretation of the homophone but relevant with respect to its other interpretation. The data suggest that people compare the main noun and the probe, and that in doing so they compute both entries of the homophone noun. No effect was found due to a variation in the delay between the presentation of the sentence and the presentation of the probe.     

The perception of clicks in linguistic and nonlinguistic messages

Three experiments were run analyzing Ss’ abilities to locate clicks in auditory messages. In Expenment 1 it was found that the intonation pattern of the linguistic message largely determined the types of errors Ss made in click placement; syntactic factors were only marginally effective and semantic factors were effectively irrelevant. In addition to these linguistic factors, there was evidence that such nonlinguistic factors as attention, memory, and response biases were contributing to the data. Experiment 2 supplied additional evidence favoring these nonlinguistic factors. Experiment 3 was a mock-up of Experiment 2 except that broad-band white noise was used as the primary message instead of English sentences. The patterns of placement errors Ss made in these several experiments (especially Experiments 2 and 3) were sufficiently similar to one another to force the conclusion that nonlinguistic factors are primarily responsible for the errors Ss make in trying to locate clicks in messages. A neo-Titchenerian attention hypothesis based upon the law of prior entry was proposed to account for the data.     

Event-related potential sensitivity to acoustic and semantic properties of terminal words in sentences.

This experiment was concerned with the effects of phonologically correct masking on the electrophysiological responses to terminal words of spoken sentences differing in contextual constraint. Two event-related potential (ERP) components, the N400 and N200, were recorded to the terminal words of high and low constraint sentences in four conditions. In the Control condition, subjects (Ss) simply attended to the sentences with no explicit task instructions. In the Semantic condition, Ss were required to listen to the stimuli in order to make semantic judgements about the terminal word of each sentence. The Control+Masking condition was identical to the Control condition except for the simultaneous presentation of a masking stimulus. The Semantic+Masking condition had Ss listening to sentences in the presence of masking with the task of making semantic judgements about the terminal word of each sentence. ERPs were recorded from Fz, Cz, Pz, T3, and T4 in 10 subjects. Amplitudes of both the N200 and the N400 were sensitive to contextual constraint with larger responses elicited by the terminal words of low constraint sentences. In addition to demonstrating the co-occurrence of the N200 and N400, this experiment highlighted a functional separation between the two components. Masking had no statistically significant effect on N200 latency but N400 latency was delayed in the masked conditions relative to those in the unmasked conditions. It is proposed that the N200 and N400 are manifestations of two different processes; the N200 reflects the acoustic/phonological processing of the terminal word while the N400 reflects the cognitive/linguistic processing. The relationship between the N200 recorded in this experiment and the discrimination N200 is discussed.     

The Bransford-Franks linear effect: Integration or artifact?

It was hypothesized that the Bransford-Franks linear effect is an artifact of the method of presentation of stimulus sentences and is unrelated to semantic processes. Ss were given sentences containing the same information in one of two ways. In a control condition, which was identical to the procedure used in earlier research, overlapping combinations of ideas were presented during learning and recognition; in an experimental condition, ideas were presented one at a time. Results demonstrated that one-idea sentences received significantly higher recognition confidence ratings in the experimental condition, thus supporting the artifact interpretation. It was proposed that Ss assign recognition confidence ratings based on the probability that a sentence containing a certain number of ideas could have occurred in acquisition.     

Decision processes during sentence comprehension: Effects of surface structure reconsidered

Previous experiments have yielded conflicting results concerning the utility of relative pronouns as cues to the deep structure underlying self-embedded (SE) sentences. The present study investigated the usefulness of this cue. At the same time, it compared the relative sensitivity of two measures of sentence-comprehension difficulty: paraphrasing and phoneme monitoring. Forty Ss heard SE sentences, one-half of which had the relative pronouns deleted. The presence or absence of the relative pronouns in a particular sentence was counterbalanced across groups. The Ss also heard right-branching sentences. It was found that presence of the relative pronouns led to significantly faster phoneme monitoring times and to marginally better paraphrasing (i.e., to better comprehension) in the SE sentences. It was concluded that relative pronouns are effective cues and, further, that the phoneme monitoring technique is a better index of comprehension difficulty than the paraphrasing technique. Reasons for the previous experimental discrepancies were discussed.     

The abstraction of linguistic ideas in “meaningless” sentences

It was proposed that the Bransford and Franks linear effect is unrelated to semantic processes and will, therefore, occur even when “meaningless” sentences (i.e., sentences containing nonsense instead of meaningful content words) are employed. Within the Bransford and Franks format, Ss were given either the meaningless sentences or control sentences. Results showed a significant linear effect for the meaningless sentences. Furthermore, although the slope of the effect for meaningless sentences was flatter than that of the control, other data ruled out a semantic integration explanation based on the availability of semantic information contained in sentence structure. A simple guessing strategy hypothesis was offered to account for the linear effect.     

The underlying structures of sentences are the primary units of immediate speech processing

Two studies of the subjective location of clicks in spoken sentences indicate: (1) within-clause phrase structure boundaries do not significantly affect the segmentation of spoken sentences; (2) divisions between underlying structure sentences determine segmentation even in the absence of corresponding explicit clause divisions in the surface phrase structure. These results support a model of speech processing according to which listeners actively segment and organize spoken sequences into potential underlying syntactic structures.     

The effect of cleft sentence structures on attention

Summary The frequency of correct recognition of agent and patient concepts as a function of agent and patient cleft sentence structures was studied. It was expected that the presentation of agent sentences would lead to a better recognition of the concept of agent than of the concept of patient. Such a different recognition performance should, however, not be found under the influence of patient sentences. These expectations were studied in an experiment in which the Ss first saw a sentence which was followed after 5 seconds by a pictorial representation of the agent and patient concepts. After the subjects had seen 12 such sentence-picture-pairs they were shown each concept individually, together with three distractors in a recognition test. The expectations were confirmed by the results.Our thanks are due to T.E. Scott M.A. for help in preparing the English text.     

Biasing thematic contexts for ambiguous sentences in a dichotic listening experiment

In an experiment derived from Lackner and Garrett (1972) 80 subjects were given a dichotic listening task where they were presented with ambiguous sentences to an attended ear and disambiguating sentences to the other, unattended, ear. Each of the sentences was preceded by a thematic context that was biased for one meaning of the ambiguous sentence. In one-half of the instances the contexts biased a meaning consistent with that of the disambiguating sentence, and in the remaining one-half they biased the meaning of the ambiguous sentence in a way that was inconsistent with the meaning of the ambiguous sentence. The meanings of the ambiguous sentences the subjects perceived tended to be those that were consistent with the biasing context, even when that meaning was inconsistent with the meaning of the disambiguating sentence. Therefore, when ambiguous sentences are preceded by a thematic context, a single-reading explanation of the processing is more appropriate than a multiple-reading explanation.