Semantic focus and sentence comprehension.

Reaction time to detect a phoneme target in a sentence was found to be faster when the word in which the target occurred formed part of the semantic focus of the sentence. Focus was determined by asking a question before the sentence; that part of the sentence which comprised the answer to the sentence was assumed to be focussed. This procedure made it possible to vary position offocus within the sentence while holding all acoustic aspects of the sentence itself constant. It is argued that sentence understanding is facilitated by rapid identification of focussed information. Since focussed words are usually accented, it is further argued that the active search for accented words demonstrated in previous research should be interpreted as a search for seman tic focus. The meanings of individual words and the combinatorial information provided by syntax do not always constitute the entire meaning of a sentence. The relationship of a sentence to its context is indicated, for example, by the focus structure of the sentence, where the focus is that information which is new and unrecoverable from preceding discourse. Speakers commonly indicate focus by assigning primary sentence accent to focussed words. Thus the new information in (1) is that it was the man on the corner, not some other man, who wore the blue hat, whereas in (2) the new information is that the hat was blue and not some other color: (1) The man on the CORNER was wearing the blue hat. (2) The man on the corner was wearing the BLUE hat. Although accenting a word is the most usual way of expressing focus, various syntactic devices can produce the same effect, for instance the topicalisation of “blue” in (3): *This research was supported by N. I. H. grant number 5-ROI-HD05168-05 to Jerry A. Fodor and Merrill F. Garrett. The authors are grateful for comments from Dianne Bradley, David Fay, Merrill Garrett, David Swinney, Virginia Valian and Edward Walker. Our thanks also to Dumont Billings, who ran the experiment, and to Edward Walker for providing the ALICE statistical analysis program. Reprint requests should be addressed to the first author at the Laboratory of Experimental Psychology, University of Sussex, Brighton, BNl 90G, England. 50 Anne Cutler and Jerry A. Fodor (3) Blue, the hat was, that the man on the corner was wearing. Linguists have devoted considerable attention to the manifestations of focus (see e.g., Halliday (1967) Jackendoff (1972)). Psycholinguistic studies of comprehension have not, so far, explicitly addressed the question of how hearers identify the focus of a sentence. Nevertheless, some indirect evidence can be adduced; focus is very often expressed as accent, and the processing of a sentence’s suprasegmental structure (accent and intonation contour) has been extensively investigated. The results of a series of phoneme-monitoring studies of accent appear to offer a starting-point for investigating the comprehension of focus. In the phoneme-monitoring task, subjects are asked to comprehend sentences and at the same time to listen for the occurrence within them of a word beginning with a specified target sound. Reaction times to the target sound in this task are faster if the target begins a word which bears accent (sentence stress) than if it begins a non-accented word (Shields, McHugh and Martin, 1974; Cutler and Foss, 1977). Since accented words are acoustically clearer in several ways (Lehiste, 1970) it might be argued that this result reflects merely the greater ease of identifying acoustically clearer words. Cutler (1976a), however, concluded that at least part of the reaction time advantage of accented targets was due to an active search for accented words on the part of the listener. In Cutler’s experiment each sentence was produced in two suprasegmental versions, one in which the target-bearing word was accented and one in which it was not. The target-bearing word was then spliced out of each version and replaced by identical copies of the same word taken from a third, fairly monotone rendition of the same sentence. Thus the target-bearing word was acoustically identical in the two versions of each sentence to be compared, but the suprasegmental context in which it occurred differed greatly from one version to another: in one version the intonation of the early part of the sentence was consistent with accent falling on the target-bearing word, in the other version it was consistent with accent falling elsewhere. Reaction times were significantly faster to targets which occurred in positions which should have borne accent than to targets which occurred where no accent was predicted. Since the acoustic clarity of the target-bearing word did not differ for the two versions, the reaction time difference in this experiment must reflect the suprasegmental differences in the part of the sentence which preceded the target. In other words, cues in the intonation contour must have enabled listeners to direct their attention to a part of the sentence where accent was about to occur. The function of sentence accent is, above all, to indicate new information. It is clear that locating accented words as quickly as possible could, therefore, Semantic focus and sentence comprehension 5 1 be a useful strategy in sentence comprehension. Once the accented words are identified, the hearer knows which part of the sentence is new information, i.e., which part is most essential for grasping the speaker’s message. Thus the reaction time advantage of accented words in phoneme-monitoring seemed to us to be due at least in part to the fact that accent expresses focus. To test this explanation, we designed the present experiment. If the reason that accent facilitates phoneme-monitoring reaction times is because it indicates focus, then, we predicted, a similar reaction time advantage for focussed words should be obtainable when focus is indicated by non-suprasegmental cues. The main design problem with such an experiment lies in the very premise on which it is based, namely, that sentence focus and accent coincide; since accent affects phoneme-monitoring reaction times, to demonstrate an analogeous effect for focus we must remove the confounding with accent, i.e., keep the acoustic representation of the sentence constant across changes in focus. However, as we have mentioned, accent is not the only method of focussing a word in a sentence.’ In this experiment focus was manipulated by preceding the sentence with a question, the answer to which was the word to be focussed. Thus when (4) is preceded by (5) it can be said to have a focus structure identical to that of (1 ), whereas preceding it by (6) will produce the focus structure of (2): (4) (5) (6) The man on the corner was wearing the blue hat. Which man was wearing the hat? What hat was the man wearing? In the present experiment, a set of sentences similar to (4) was presented to subjects. Half the subjects heard the sentence preceded by (5) half heard it preceded by (6). The experimental tapes were spliced together from separate recordings of sentences and questions, so that the monitored sentence (4) was actually acoustically identical in each presentation. Because acoustic cues in the sentence could favor one monitoring location over another, each sentence contained two possible phoneme targets; within each of the two question groups, half the subjects listened for one target, half for the other. Thus ’ These methods include topicalisation, clefting (“It is the man who is wearing the hat”) and pseudoclefting (“The one who is wearing the hat is the man”), all of which are less suitable than the method actually adopted since they involve variations of the surface structure of the sentences to be compared. Allen and O’Shaughnessy (forthcoming) report that various devices used to indicate sentence focus-clefting, pseudo-clefting, topicalisation, preposed question-produce reliable and essentially similar effects on the suprasegmental contour; fundamental frequency accent occurs in each case on the element which is focussed. (In the present experiment these explicit cues were of course not present since the questions were produced separately from the sentences). 52 Anne Cutler and Jerry A. Fodor in (4) half the subjects listened for a word beginning with /k/, half listened for a word beginning with /b/. If phoneme-monitoring reaction time were indeed to prove sensitive to focus variation in the manner suggested by the results from the accent experiments, then it was expected that subjects monitoring in (4) for the /k/ sound on the word “corner” would respond faster if (4) was preceded by (5) than if it was preceded by (6), while subjects monitoring in the same sentence for the /b/ sound on “blue” would produce faster reaction times if (4) was preceded by (6) than if it was preceded by (5). Thus we predicted an interaction between the two variables of Focus Position and Target Position, such that reaction times to earlier targets would be faster when the preceding question focussed upon the earlier part of the sentence, whereas reaction times to later targets would be faster when focus was upon the later part of the sentence. On the basis of other findings, a main effect of Target Position was also predicted. It is a reliable finding in phoneme-monitoring experiments that targets near the end of a sentence produce faster reaction times than targets near the beginning of a sentence (Foss, 1969; Shields, McHugh and Martin, 1974). It has been argued (Cutler, 1976b) that this effect reflects the construction and testing of semantic hypotheses during sentence comprehension, with a greater probability of correct hypotheses being constructed as the sentence progresses towards its end.