Outline and discuss any evidence that Relevance (or QUD) is computed incrementally

It has been widely accepted that the comprehension of a sentence involves the process of both interpreting sentences meaning and understanding how that sentence is relevant to the context. The time-sensitive data collected from comprehension tasks suggested that these two processes are incremental and interactive （Tian， 2016）. In this paper， we outline the ideas that relevance is achieved incrementally， and discuss a study of negation processing to support this idea.

“Question under discussion” （QUDs; see Ginzburg， in press） is a term to describe the relevance achieving process. It characterizes relevance as a set of semantic inquiries that require to be solved. According to Roberts （2012）， the concept is developed based on his idea of information structure， a structure on pursuing inquiry in discourse and acquiring information that inquiry yields. The basic idea is that language users represent parts of utterances as questions， and whatever utterances they encounter， they tend to project the most prominent QUDs of it to achieve relevance. The property of QUDs indicates its computation process to be incremental. The prominence of QUDs change as discourse unfolds， and can be adjusted by a mechanism of accommodation. This kind of adjustment is in response to features of the utterances offered. Whats more， QUDs accommodation happens in parallel with the content processing， and the likely QUD of an utterance is anticipatory.

The current study of negation processing provides an evidence for the incremental computation of QUD （Tian， 2016）.

For decades， it has been observed that negation processing is difficult without context since that it takes longer to process and there are more errors occur compared with the processing of positive sentences （Wason， 1965; Clark&Chase， 1972）. Two kinds of approaches rise to account for the delay of negation processing. Truth-functional approaches assume that the cost of negation processing is caused by extra step. By contrast， ‘contextualist approaches assume that the delay is caused by the need for a context for negation.

In previous research of negation processing， many studies point out that when interpreting negative sentences， participants tend to represent positive argument of negation first. Some sentence verification experiments suggest that when interpreting a negative sentence， people first represent the negated information， and then reject it. Kaup and colleagues established a rejection model based on truth-functional approach to account for the representation of positive argument. A visual probe recognition task was conducted and the results show that given a negative sentence （eg. The bird is not flying）， within 250ms， participants respond faster to a picture consistent with the positive argument of negative sentence （a flying bird） than to a target picture （a resting bird）. Therefore the conclusion was made that negation is represented with a rejection model that involves two-step simulation： first simulate the negated information and then reject it.

However， it is not always the case that the positive argument is firstly and necessarily represented in negation processing. Therefore the truth-functional approach fails to account for some cases. Tian， Breheny and Ferguson? provided such a piece of evidence that the representation of positive argument doesnt occur in interpreting cleft negative sentences （It is Tom that has not iron his brothers shirt）.

Here come questions： why does the positive argument sometimes occur in negation processing while sometimes not？ What activates the representation of negated information during the processing？ The ‘contextualists approaches provide an alternative account that is based on the incremental property of QUD accommodation.

The dynamic pragmatic account suggests that it is QUD accommodation triggers the positive representation. Based on the view that negation requires specific contextual motivation and negative question are marked， given a negative sentence without any context， participants are assumed to accommodate the more frequent positive question querying either a positive or negative argument as the prominent QUD. For example， when they hear a sentence like， “Tom hasnt iron his brothers shirt”， participant tend to project the most prominent QUD asking whether Tom has iron his brothers shirt， and this is a positive question. This is because the major property of negation is denial and rejection. However， when the given sentence has a negative question as its prominent QUD， then there will be no positive representation in sentence interpreting process. Cleft negative sentence is an example of this kind. For example， when participants hear a sentence like， “Its Tom that hasnt iron his brothers shirt”， participant tend to accommodate a QUD querying who hasnt iron his brothers shirt， and in this case no positive argument is presented.

In order to test the exact point that QUDs accommodation occur and whether it happen incrementally， Tian et al. conducted a visual world eye-tracking task to compare the time course of negative sentences processing and their positive counterparts representations. According to incremental QUD accommodation account， the prediction is that the results of time-course task of projecting QUDs should be the same in either positive or negative condition. The experiment has the similar two by two material sentences format classified by polarity and cleft-ness as the previous one， but the visual contexts of two propositions （consistent or not consistent with the given sentence） are given one second before participant hearing the audio stimuli to eliminate the disadvantage of negative states inferences. For example， for the group of 4 sentences of “turn on the TV”， the visual context shows two states of TV， either on or off. Together with a person and two distractors， there are totally 5 items on each visual scene. Participants are instructed to simply look at the visual scene while listening， and their eye movements for each trial are recorded.

The eye-movement analysis of the study focuses on the comparison between the percentage of participants looks to the target and competitor from the onset of the verb to the offset of the final noun. The looks are calculated as ‘target advantage， which directly show the comparison. The average target advantage scores are calculated by person and by item. Results show that in simple sentences， participants show a bias to target immediately after the offset of verb in positive sentences but before the onset of noun in negative sentences. In other word， in the duration between the offset of verb and onset of noun， participants fixate the target earlier in positive sentences than negative sentences. However， in terms of cleft sentences， participants show no significant difference in processing positive sentences and negative ones， which indicates that polarity in cleft sentences doesnt influence the processing time.

An analysis of time-course is also conducted to investigate the exact time when target bias was developed in both sentence types. The time that starts from the offset of the verb to the end of a sentence is divided into several 100ms time slices， and the target advantage scores for each slice are calculated. In the first 5 slices after the offset of the verb， simple sentences show great difference between positive and negative while cleft positive and negative sentences show nearly the same target advantage scores. After the slice 6， both sentences type show different scores between positive and negative.

The analysis of eye-movement and time-course provides evidences to support the idea that QUDs accommodation is incremental and happens in parallel with the content processing. First， the results for cleft negative sentences indicates that the meaning of negation is integrated incrementally. When processing cleft negative sentences， participants fixate the target only 600ms after the verb， which means that after hearing the verb they immediately combine the negation information and the verb together to direct their looking preference. Second， there is no delay in cleft negative compared with cleft positive， which indicates that the prominent QUDs for cleft negative are negative questions as well. This is because during the incremental processing， when hearing “has”/ “hasnt”， participants using this cue to recognize the sentence form （cleft sentence）， and then start to predict the sentence meaning and the prominent QUD at the same time incrementally as linguistic input unfold.

Language is processed incrementally， and this process involves not only the content processing but also the relevance achievement. QUD accommodation serves as a way to achieve relevance， and the computation of QUD is incremental as well. When hearer process an utterance， he represents it as a set of queries waiting for resolution， among which the most prominent one is the question that both speaker and hearer agree to be the focus in a broader discourse. QUD accommodation is updated as discourse unfolds， and is parallel with the content interpreting. In negation research， this dynamic pragmatic account is incompatible with rejection account. QUD accommodation explains the occasional representation of positive argument in negation processing and denies the necessity of “rejection” process. It is the incremental QUD accommodation that allows the negation in cleft sentences to be processed as fast as the positive cleft sentences.

References：

[1]Ginzburg， J. The Interactive Stance[M]. CSLI： Center. Oxford University Press，2012.

[2]Huddleston， R.， & Pullum， G. K. The Cambridge Grammar of the English Language[J].2002.

【作者簡(jiǎn)介】張苒（1982.12-），女，河北館陶人，河北師范大學(xué)附屬民族學(xué)院，講師，研究方向：英語(yǔ)語(yǔ)言文學(xué)。