Emmanuel Chemla

Experimental Evidence for Embedded Scalar Implicatures

Scalar implicatures are traditionally viewed as pragmatic inferences which result from a reasoning about speakers’ communicative intentions (Grice 1989). This view has been challenged in recent years by theories which propose that scalar implicatures are a grammatical phenomenon. Such theories claim that scalar implicatures can be computed in embedded positions and enter into the recursive computation of meaning—something that is not expected under the traditional, pragmatic view. Recently, Geurts and Pouscoulous (2009) presented an experimental study in which embedded scalar implicatures were not detected. Using a novel version of the truth value judgment tasks, we provide evidence that subjects sometimes compute embedded scalar implicatures.

Categorizing words using ‘frequent frames’: what cross-linguistic analyses reveal about distributional acquisition strategies

Mintz (2003) described a distributional environment called a frame, defined as the co-occurrence of two context words with one intervening target word. Analyses of English child-directed speech showed that words that fell within any frequently occurring frame consistently belonged to the same grammatical category (e.g. noun, verb, adjective, etc.). In this paper, we first generalize this result to French, a language in which the function word system allows patterns that are potentially detrimental to a frame-based analysis procedure. Second, we show that the discontinuity of the chosen environments (i.e. the fact that target words are framed by the context words) is crucial for the mechanism to be efficient. This property might be relevant for any computational approach to grammatical categorization. Finally, we investigate a recursive application of the procedure and observe that the categorization is paradoxically worse when context elements are categories rather than actual lexical items. Item-specificity is thus also a core computational principle for this type of algorithm. Our analysis, along with results from behavioural studies (Gómez, 2002; Gómez and Maye, 2005; Mintz, 2006), provides strong support for frames as a basis for the acquisition of grammatical categories by infants. Discontinuity and item-specificity appear to be crucial features.

An Epistemic Step for Anti-Presuppositions

Sentence (1) strongly suggests that the speaker does not have a sister: (1) John believes that I have a sister. a. Alternative: John knows that I have a sister. b. Actual inference: The speaker does not have a sister. c. Predicted inference: It is not common belief that the speaker has a sister. According to Heim (1991), Percus (2006), and Sauerland (2006), this inference should follow from the comparison of (1) to (1a). However, such an analysis would only predict a very weak implicature: it is not common belief that the speaker has a sister. I propose to strengthen this prediction by two means. First, I rely on a precise understanding of the modern Stalnakerian view of presuppositions and common ground (Stalnaker 1998, 2002; von Fintel 2000; Schlenker 2006). Second, I argue that this inference depends on contextual factors. More precisely, I show that the Competence Assumption (see Spector 2003; van Rooij & Schulz 2004; Sauerland 2004) necessary to obtain secondary scalar implicatures should be supplemented with an Authority Assumption. I motivate this additional assumption on independent empirical grounds. Finally, I show how my proposal accounts for a wide variety of inferences with fine variations governed by (i) contextual differences and (ii) specific properties of the presupposition triggers involved.

Cross-Situational Word Learning in the Right Situations.

Upon hearing a novel word, language learners must identify its correct meaning from a diverse set of situationally relevant options. Such referential ambiguity could be reduced through repetitive exposure to the novel word across diverging learning situations, a learning mechanism referred to as cross-situational learning. Previous research has focused on the amount of information learners carry over from 1 learning instance to the next. In the present article, we investigate how context can modulate the learning strategy and its efficiency. Results from 4 cross-situational learning experiments with adults suggest the following: (a) Learners encode more than the specific hypotheses they form about the meaning of a word, providing evidence against the recent view referred to as single hypothesis testing. (b) Learning is faster when learning situations consistently contain members from a given group, regardless of whether this group is a semantically coherent group (e.g., animals) or induced through repetition (objects being presented together repetitively, just like a fork and a door may occur together repetitively in a kitchen). (c) Learners are subject to memory illusions, in a way that suggests that the learning situation itself appears to be encoded in memory during learning. Overall, our findings demonstrate that realistic contexts (such as the situation in which a given word has occurred; e.g., in the zoo or in the kitchen) help learners retrieve or discard potential referents for a word, because such contexts can be memorized and associated with a to-be-learned word.

Phantom readings: the case of modified numerals

We investigate the mechanisms proposed in formal semantics to account for the ambiguity generated by simple numerical expressions (e.g., ‘three students’). We explain how these mechanisms, when applied to more complex numerical expressions such as ‘between n and m’ (e.g., ‘between three and five students’), predict a surprising ambiguity between a doubly bounded (e.g., ‘at least three and at most five students’) and a lower-bounded reading (e.g., ‘at least three students’). While the lower-bounded reading is not detectable intuitively, results from three offline experiments and a response time study provide evidence in favour of its existence. Our contribution is twofold. On the experimental side, we present two psycholinguistic methods powerful enough to detect what we call phantom readings, i.e., readings that do not seem to have consequences for actual interpretation, but have detectable effects on processing. On the theoretical side, we show that certain semantic mechanisms that might be thought to overgenerate are in fact vindicated, since they are able to predict certain processing facts that would otherwise remain mysterious. We discuss how these results illustrate the need for a strong integration of formal semantics and psycholinguistic approaches.

What Homophones Say about Words

The number of potential meanings for a new word is astronomic. To make the word-learning problem tractable, one must restrict the hypothesis space. To do so, current word learning accounts often incorporate constraints about cognition or about the mature lexicon directly in the learning device. We are concerned with the convexity constraint, which holds that concepts (privileged sets of entities that we think of as “coherent”) do not have gaps (if A and B belong to a concept, so does any entity “between” A and B). To leverage from it a linguistic constraint, learning algorithms have percolated this constraint from concepts, to word forms: some algorithms rely on the possibility that word forms are associated with convex sets of objects. Yet this does have to be the case: homophones are word forms associated with two separate words and meanings. Two sets of experiments show that when evidence suggests that a novel label is associated with a disjoint (non-convex) set of objects, either a) because there is a gap in conceptual space between the learning exemplars for a given word or b) because of the intervention of other lexical items in that gap, adults prefer to postulate homophony, where a single word form is associated with two separate words and meanings, rather than inferring that the word could have a disjunctive, discontinuous meaning. These results about homophony must be integrated to current word learning algorithms. We conclude by arguing for a weaker specialization of word learning algorithms, which too often could miss important constraints by focusing on a restricted empirical basis (e.g., non-homophonous content words).

Revealing abstract semantic mechanisms through priming: The distributive/collective contrast

Sentences such as The bags are light allow both collective (they are light together) and distributive interpretations (each bag is light). We report the results of two experiments showing that this collective/distributive contrast gives rise to priming effects. These findings suggest that collective and distributive readings involve different interpretative mechanisms, which are at play during real comprehension and can be targeted by priming, independently of the specific verification strategy associated with each interpretation.

Cross-linguistic regularities and learner biases reflect "core" mechanics.

Recent research in infant cognition and adult vision suggests that the mechanical object relationships may be more salient and naturally attention grabbing than similar but non-mechanical relationships. Here we examine two novel sources of evidence from language related to this hypothesis. In Experiments 1 and 2, we show that adults preferentially infer that the meaning of a novel preposition refers to a mechanical as opposed to a non-mechanical relationship. Experiments 3 and 4 examine cross-linguistic adpositions obtained on a large scale from machines or from experts, respectively. While these methods differ in the ease of data collection relative to the reliability of the data, their results converge: we find that across a range of diverse and historically unrelated languages, adpositions (such as prepositions) referring to the mechanical relationships of containment (e.g “in”) and support (e.g. “on”) are systematically shorter than closely matched but not mechanical words such as “behind,” “beside,” “above,” “over,” “out,” and “off.” These results first suggest that languages regularly contain traces of core knowledge representations and that cross-linguistic regularities can therefore be a useful and easily accessible form of information that bears on the foundations of non-linguistic thought.

An Attentional Effect of Musical Metrical Structure

Theories of metrical structure postulate the existence of several degrees of beat strength. While previous work has clearly established that humans are sensitive to the distinction between strong beats and weak ones, there is little evidence for a more fine grained distinction between intermediate levels. Here, we present experimental data showing that attention can be allocated to an intermediate level of beat strength. Comparing the effects of short exposures to 6/8 and 3/4 metrical structures on a tone detection task, we observe that subjects respond differently to beats of intermediate strength than to weak beats.