Effects of context on the processing of adversative and comparative constructions

Publication
TitleEffects of context on the processing of adversative and comparative constructions
Publication TypeJournal Article
Année de publication2014
AuthorsWinterstein, G., Ellsiepen E., Jayez J., Hemforth B., Winterstein G., Ellsiepen E., Jayez J., & Hemforth B.
JournalCUNY
Date Published2014
Keywordslinguistics, SHS.LANGUE Humanities and Social Sciences/Linguistics
Abstract

This work focuses on the combination of adversative connectives (e.g. but) with com- parative constructions. The basic observation is that, when introduced by an adversative, a comparative expression can only indicate inferiority; mentioning superiority leads to infelicity (1). Replacing but by and makes the contrast vanish, confirming that but is partly responsible for it. 1. a. The Friday exam was difficult, but less difficult than the Tuesday exam. b. # The Friday exam was difficult, but more difficult than the Tuesday exam. Intuitively, the contrast is less sharp if those examples are used in a facilitating context. For (1b) such a context could be that a teacher has two classes and is supposed to give them equally difficult exams. Asking whether the teacher achieved this balance makes (1b) sound more ac- ceptable as an answer by making explicit the opposition necessary for interpreting but. Theories of but deal with this picture in different ways. Relevance Theory (RT) 1 con- siders that a but-conjunction requires an element that is accessible from the first conjunct and world-knowledge and that gets denied by the second. Without context, it is assumed that this element cannot be found in (1b). In context, the use of but is perfectly felicitous since this ele- ment is explicitly made accessible. In Bayesian Argumentation Theory (BAT) 2, the contrast in (1) is accounted for on semantic grounds. The infelicity of (1b) is explained by considering that both conjuncts are lower-bounding expressions and thus that their probabilistic effects are simi- lar: their assertions modify degrees of belief in comparable ways, hampering the abduction of the opposition necessary for but. Therefore, even with a context, (1b) is expected to differ from (1a) since the semantic argumentative clash remains present, even if context overrides it. In order to test the differences between RT and BAT, we conducted a region-by-region self-paced reading experiment immediately followed by an acceptability judgment study in French. The twelve experimental items consisted of a test sentence and a context designed to rise the expectation of some property being equal for two entities (as in (1)). Test sentences were of the same form as in (1), in the more condition the comparison was formed with mais plus + adjective, while in the less condition it was formed with mais moins + adjective. Half of the participants, the no-context group, were presented the test sentence in isolation, while the other half, the context group, saw the test sentence preceded by the facilitating context. After completing the reading part, participants were asked to judge the sentences’ acceptability on a 9 point scale. For the context group, sentences were again preceded by the context. Analysis of the acceptability ratings using mixed models indicated both, a main effect of condition (χ(1) = 8.14,p < .01), and an interaction of condition and group (χ(1) = 4.59,p < .05). A follow-up analy- sis showed that the effect of condition was only significant in the no-context group, confirming the remedial effect of context on the acceptability of sentences like (1b). For the self-paced reading data, we analyzed residual reading times from the critical region containing the comparison, as well as the three following regions to allow the context to take effect on reading times. Condition had a significant effect in region 3 (χ(1) =4.00,p < .05), which followed the comparison object, where reading times were longer in the more condition. Here, the interaction with group was not significant. This suggests that while context affect offline measures, online difficulties arise both, with and without context, as predicted by BAT.

URLhttps://hal.archives-ouvertes.fr/halshs-01428507
DOI10.13140/RG.2.1.1099.1520