This paper describes the semantics of the Cantonese Sentence Final Particle tim1. We argue in favor of distinguishing two usages of the particle. We present acoustic, syntactic and semantic evidence supporting our claim. We formalize the meaning of the two particles using a probabilistic framework and use this formalization to argue for a pragmaticalization of the additive meaning toward the mirative meaning.

This paper deals with the Cantonese morpheme gwai2 which conveys expressive meaning when modifying a wide range of expressions: adjectives, verbs, wh-pronouns, etc. We review its empirical domain and discuss different claims of the literature concerning its dual nature as an intensifier and a mixed-expressive conveying at-issue negation. We argue that gwai2 cannot be treated as an intensifier in the usual sense, and that it does not contribute a truth-conditional negation, but rather a form of denial. We then propose a unified analysis of the morpheme based on the assumption that it indicates a negative attitude of the speaker towards its argument, notably by showing how to derive denials from this negative attitude.

This paper presents experiments which measured explicit gender biases in Japanese and English. We use clustering analysis to distinguish between different types of biases. We integrate these results in a Bayesian model of arguments from authority to predict how epistemic biases influence the convincingness of these arguments.

This paper analyzes how linguistic cues can affect the epistemic authority of an agent, i.e. their reliability and how likely other agents are likely to form beliefs based on their speech. We begin by reviewing McCready’s (2015) theory of reliability, and then turn to strategies for attempting to assume epistemic authority, focusing on those involving the use of not-at-issue content. We then show the results of two experiments which test the interaction of stereotypes about gender with epistemic authority, and how this is mediated by language use, focusing on the case of gendered pronouns. Finally, the results are explored for Bayesian views of argumentation and analyzed within McCready’s Reliability Dynamic Logic.

This paper deals with the property of obligatoriness that is often described as characteristic of the class of additive particles (e.g. the adverb 'too'). In a nutshell, these particles are said to be obligatory because omitting them in a discourse creates either an infelicity or the derivation of unwanted inferences. A common trend of analysis treats this property as boolean. The general consensus used to be that if an additive particle can be used in a discourse then it has to be used. However some examples show that this property is probably best treated as a gradient. In this work we investigate some of the conditions that affect this obligatoriness. First, based on a small experiment with corpora, we evaluate the frequency of the obligatory vs .optional uses of too. Then, having established that the optional cases are not the exception, we present the results of an experiment that evaluates the effect of ellipsis and anaphora on the obligatoriness of additive particles.

The paper details several experiments focusing on the argumentative side of only. The experiments confirm the theoretical prediction that utterances with similar informational content can behave differently from the argumentative point of view. It is proposed to adapt Bayesian approaches to argumentation by restricting the type of content they take into account to model the argumentative impact of an utterance.

This work deals with the way discourse markers combine and the interpretative effects that come out of such combinations. We focus on the case of the relations Contrast and Parallel as marked by the connectives but and too. We show that any theory of discourse interpretation has to allow for the two relations to combine. We then look at how a specific theory, SDRT, can handle these cases.

This paper describes the construction of a digital corpus containing most of the known inscriptions in Linear A.

The paper describes an algorithm for opinion classification. The system parses the dependency structure of an utterance and based on it derives its positive or negative orientation by combining the primitive polarities of elements (based on SenticNet) and taking into the effect of some specific markers. Results are comparable to the state of the art.

We use offline and online measurements to investigate how contextual information is integrated in the interpretation of argumentative elements. We focus on adversative conjunction and comparative constructions. Results show that even in the presence of contextually available remedial information, participants still exhibit online difficulties processing utterances which involve argumentative clashes.

We present a series of tests to determine whether the French preposition pour, when introducing an infinitive verb phrase, can be considered as a discourse connective. This work is part of the FDTB, a larger project which aims at annotating discourse relations in the Paris 7-French Treebank.

This work deals with sentiment analysis on a corpus of French product reviews. We first introduce the corpus and how it was built. Then we present the results of two classification tasks that aimed at automatically detecting positive, negative and neutral reviews by using various machine learning techniques. We focus on methods that make use of feature selection techniques. This is done in order to facilitate the interpretation of the models produced so as to get some insights on the relative importance of linguistic items for marking sentiment and opinion. We develop this topic by looking at the output of the selection processes on various classes of lexical items and providing an explanation of the selection in argumentative terms.

This paper looks at the way some discourse relations interact with the dimension of quantity implicatures. I focus on the relation of contrast as marked by the connective but to argue that it is preferable to assume that the way contrast is established does not take quantity implicatures into consideration. I explore the consequences of this for a set of theories of but which we refer to as being inferential.

In this work, we give a new semantics to the notion of additivity as embodied by several discourse markers and particles in French: et, de plus and d'ailleurs. The common property of these different elements is the notion of independence of their arguments. We show that existing accounts of additive particles fail to do full justice to this notion of independence, and we propose a new semantics for and that captures this notion in a Bayesian fashion. We then evaluate the applicability of this analysis to other markers in French.

This paper discusses the semantics of the connective but. Two trends of analyses are compared: ones based on a notion of formal contrast and others that are inferential. First, the formal contrast approaches are evaluated with respect to a certain number of problematic examples. I argue that they encounter insurmountable issues, and that an inferential account is needed. However, the way the inference required in those latter accounts is drawn needs to be defined in a restricted way. I propose to use the probabilistic interpretation of the notion of argumentation to carry this out.

A wide-spread view is that the distribution of too falls apart in just two cases: either too can appear and then it must appear or it is not allowed to appear and then it is forbidden. In the first case, the context supplies a proper antecedent for too, in the second case it does not. The paper argues that there are also optional uses and cases where there is a proper antecedent, but too is better avoided. The paper discusses the problems these observations pose to existing theories about the production and interpretation of additive particles like too and suggests some solutions.

After observing that the emphatic uses of the English determiner any have an effect of the possible discourse continuations, we propose to model these effects in a probabilistic framework. This move is warranted by the fact that successful probabilistic models have already been proposed both for the semantics of any and those of discourse relations.

This work deals with the semantics properties of additive presupposition triggers. In particular, we look at the question of the apparently obligatory status of these markers by showing that this property is not systematic, contrary to what has generally been supposed in the literature. To explain cases of optional additive marking, we use the hypothesis developed in (Winterstein & Zeevat, 2012) regarding the argumentative nature of additive particles.

Traditional descriptions of the adverb only usually postulate that this element includes a scalar component that imposes that its associate belongs to the lower end of a contextual scale. On the basis of a particular class of examples, I argue that these effects are not systematic and that it is possible to derive the scalar effects of only with a simple exclusive component.

This work investigates the projection problem for presuppositions using the framework of ludics for modeling dialogue. It is shown how ludics naturally integrate a series of observations on the discursive status of presuppositions and thus has the potential to shed more light on this phenomenon.

In this article, we look at the Romanian equivalents of the conjunctions and and but. Instead of having two conjunctions, like in English, Romanian has four different connectives that cover the same semantic space. One of them (iar) is close in meaning to a connective usually attested in Slavic languages (e.g. a in Russian). We show that the use of iar is nevertheless more restricted than that of a and we analyze this difference by showing that other Romanian connective are closer to connectives attested in Romance languages than those found in Slavic ones. The competition between iar and its competitors is thus not the same as for a, which explains their different distributions.

This work evaluates the possibility of applying the insights of argumentation theory to theories of discourse using game theoretical mechanisms. The main question we look at is whether it is relevant to interpret argumentative strength as a measure of the speaker's utility. We argue against this move by showing that argumentation characterizes utterances types whereas utility can only apply to actual tokens.

This article deals with the semantics of the adverb too (and its French equivalent aussi) and has two aims. First, it is shown that the semantics of too cannot be limited to a single presuppositional content. To show this I rely on contexts that verify the presupposition of too but do not allow its use. I then postulate that too encodes a discursive similarity constraint, distinct from its presuppositional content. I describe this constraint in argumentative terms.

Dissertation committee:

• Pascal Amsili (U. P7)
• Nicholas Asher (CNRS-IRIT)
• Francis Corblin (U. P4)
• Jacques Jayez (ENS-Lyon)
• Jean-Marie Marandin (CNRS-LLF)
• Henk Zeevat (U. Amsterdam)

Two treatments for the semantics of the adversative connective but are compared: the argumentative one and the based on a notion of formal contrast. It is shown that none is enough to cover all the empirical domain of the conjunction, and a hybrid semantics for the connective is proposed.

The foundational observation of this work deals with the reinforcement of conversational implicatures. It appears that Quantity implicatures must be reinforced with a discourse segment introduced by the connective \emph{but} whereas this same connective is used to cancel other inferential material. I show that this observation is not as robust as it seems by exhibiting contexts such that canceling a quantity implicature can be done by using \emph{but}. I then explain the preference to use \emph{but} in the reinforcement cases by resorting to the argumentative properties of the elements.