What did he say that for? Some contextual effects on the process of understanding a sentence

Abstract

Though it seems intuitively very probable that processing of sentences is likely to be affected by the environment - both linguistic and non- linguistic - in which the sentence is encountered, remarkably little work has been carried out to investigate the parameters of the process. While some work in recent years (for example that of Bransford and his co-workers) has amply demonstrated that what is remembered from a sentence is strongly influenced by other sentences which have to be remembered, few people have carried out investigations like those of Huttenlocher and her co-workers into how the situation can affect comprehension of a sentence. Psycholinguistic work has for the most part been directed towards the process of understanding single sentences shorn, so far as possible, of any ties with either other sentences or the real world. This unfortunate occurrence seems to have been largely due to the overwhelming influence of transformational grammar, which encourages one to believe that processing of sentences constituting part of a text is much the same as processing of single sentences. At any rate the core processes would appear to be the same on this account. In the present work an alternative theory of grammar is utilised. This theory, called systemic grammar, places great emphasis on the information structure of a sentence and hence, implicitly, on the relation between sentences and the context in which the sentence is encountered. The present work concentrates pre-eminently on the influence of other sentences on the processing of a particular sentence - rather than on the influence of the accompanying situation. However, much is said on the subject of Huttenlocher's work. The work starts with a highly selective review, heavily reliant on Clark (1974), of relevant literature. The review concludes that the bulk of the work reported in the literature lacks a coherent theory within which to conceptualise contextual influences on sentence processing. It is also suggested that a number of effects reported in the literature may be heavily dependent on contextual influences. Systemic grammar is put forward as a theory which might enable us to conceptualise some of the factors relevant to understanding sentences in context. A brief outline of systemic theory is given. Following this, nine experiments are reported on a variety of traditional effects, but in all cases manipulating whether the target sentences occur as part of a text or not. In addition variation of such cohesive devices as the use of pronouns and the definite article also occurs. The first experiment involves the systematic manipulation of definiteness marking, syntax, relational term and presence of text. A verification task is used in which the sentence precedes the picture with separate measures of comprehension (how long subjects choose to have the target sentence exposed) and verification (how long it takes them to respond when they see the picture). Interesting results occur in both sets of data but the main findings are : firstly, that the traditional lexical marking effect only occurs with single sentence presentation - when embedded in a longer text there is no difference between reaction times to the marked and unmarked words; secondly, although the marked syntactic form (in this case with the locative phrase before rather than after the copula) tends to be harder to understand this effect almost entirely disappears when the sentence is in a context and the topic of the paragraph is the theme (first noun) of the sentence - with unmarked syntax reactions are quicker if the topic is the second noun, but neither of these effects occur, of course, in the no text case where there is no topic; thirdly the marked lexical form was responded to faster if the two nominals were marked differently for definiteness, whereas the unmarked form tended to be responded to more rapidly if both nominals were similarly marked. The first and last results were explained as due to a "good reason" interpretation of marking in which marking is considered subordinate to topicalisation choices and the marked form does not convey additional information if it can be seen to have been chosen for that reason. The second result received a related explanation, though with a fuller analysis of the role of this marked syntactic form. Experiments 2 and 3 investigate precisely the same phenomena as Experiment 1. The first of these requires subjects to write down a series of sentences of the same sort as those presented to subjects in Experiment 1, to describe a series of pictures presented to them. Various constraints are built into the task in order to encourage them to produce a wide variety of responses. The frequency of different forms is similar to what one might expect from Experiment 1 given the assumption that reaction times and production frequencies are inversely related. Similar effects were observed to those in Experiment 1 with the exception that definiteness marking was seen to be of much greater importance in this experiment. This last result was also replicated in Experiment 3 - a much less constrained study in which subjects described pictures orally. Classifying responses on the basis of a large number of criteria this study demonstrated that very few of the possible responses occurred. However many more occurred with unmarked than with marked syntax - supporting the interpretation of the marked option as not in itself more complex, but rather with more complex entry conditions (selected in a narrower set of circumstances). This study also demonstrated some interesting differences in the patterns of use of pronouns and the definite article. The next two experiments follow up this last point by investigating differences between pronouns and other means of cross-referring in terms of reaction times. It is clear from these two experiments that pronouns do not simply speed up comprehension relative to other methods. The effect seems to depend upon several factors including the information structure of the sentence. The fifth experiment used the three term series problem to examine the use of pronouns, lexical marking and Huttenlocher's result that the second premise is easier to understand if the new item is first in that premise. Reaction times to the first premise, the second premise and the question were measured separately. Huttenlocher's effect was greatly enhanced by the use of a pronoun in the second premise to cross refer to the first premise. This was interpreted as being due to pronouns making clear the new and old information parts of the sentence and so enabling subjects to take advantage of the fact that their primary focus of interest when reading the second premise - namely the third object - is referred to by the more prominent theme, something which is more usually reserved for old information. A second factor influencing processing of sentences with pronouns in them is whether the pronoun in the second premise refers to the same object as the subject or object of the first premise. Subjects respond more rapidly if it is co-referential with the subject. Experiment 3 demonstrated that this is also the more common occurrence in free descriptions. Other results in this experiment provided more support for the interpretation of lexical marking in terms of a good reason principle: there being a strong effect of marking of the first premise (where it is hard to see any topicalisation reason for choosing it) but no straightforward effect of marking of the second premise. Furthermore marked questions do not take longer to process than unmarked - in fact, thanks probably to an interaction, they actually take significantly less time. The next three experiments again involve verification but here the presentation of sentences was experimenter controlled and oral. Reaction times were again used but the measure taken was a complex comprehension/verification one. This measure was supplemented by a measure of the number of fixations subjects made in scanning the picture. This set of data was analysed in much the same way as the reaction time results. Experiments 6 and 7 involved successive presentation of sentence and picture (in that order), while Experiment 8 Involved simultaneous presentation. On the whole the latter was more successful but this may have been because more complex pictures were used. The fixation data, though producing a number of apparently reliable results, did not produce results which bore any clear relationship to the reaction time data and evidence to the contrary reported by Hall (1975) is therefore called into question. However the reaction time data as a whole are not very clear in these three experiments. Only in Experiment 8 in which passives are shown to be easier to understand when the theme is previously mentioned, and actives when the theme is not previously mentioned (this is true, of course, only for the context condition) are there any very clear results. In Experiments 6 and 7 on the other hand, it does seem that passives are only harder to understand than actives if they are false, but Experiment 8 only shows a simple effect of truth value. Effects of context in Experiments 6 and 7 are not large - possibly this is due to the delay between presentation of the sentence and the taking of any measure. The final experiment again used the text manipulation and like Experiment 8 presented sentence and picture simultaneously, but the sentence was a question which had to be answered rather than a statement to be verified. Questions differed in whether the noun preceded or succeeded the main verb, in voice, and in whether the noun was previously mentioned or not. On the whole results approximated quite closely to what one might expect from corresponding declaratives and a functional interpretation of the systemic options involved. A feature of both this experiment and the previous one is the use of two sets of reaction time data: data from the onset of the question to the onset of the answer and data from the offset of the question to the onset of the answer. On the first analysis passives take consistently longer to process, but on the second they are, if anything, processed faster. A final chapter summarises some of the major results and compares both the experimental methods and the measures used in the various experiments. On the whole the conclusion is that sentence by sentence presentation for subject-controlled durations is the most satisfactory method. The gross measure of number of fixations is not seen as a useful one, though it is suggested that with simultaneous presentation of sentence and picture a moment by moment comparison of the sentence with what the subject Is fixating may be of interest. The main substantive contributions of the present work are seen as: (1) further evidence that the canonical form view of sentence processing is unhelpful, (2) a good deal of support for the "good reason" approach to both lexical and grammatical marking which explains the greater difficulty of marked forms as due, not to the fact that they are themselves more complex, but to the fact that the reasons for selecting them (entry conditions) are more complex, (3) some preliminary evidence of the effects on RT of a handful of cohesive devices among them the use of definiteness marking, pronouns, lexical marking, the passive voice and certain other marked syntactic configurations.