The quantifier “some” often elicits a scalar implicature during comprehension: “Some of today's letters have checks inside” is often interpreted to mean that not all of today's letters have checks inside. In previous work, Goodman and Stuhlmüller (G&S) proposed a model that predicts that this implicature should depend on the speaker's knowledgeability: If the speaker has only examined some of the available letters (e.g., two of three letters), people are less likely to infer that “some” implies “not all” than if (...) the speaker has examined all of the available letters. G&S also provided behavioral evidence in support of their model. In this paper, we first show that a simple extension of G&S's model (1) predicts G&S's knowledgeability effects, and in addition, (2) predicts that the knowledgeability effect will be reduced when the speaker's usage indicates numeral alternatives are available. We tested the new model's predictions in four preregistered experiments. All experiments supported the first model prediction, replicating G&S's finding of a main effect of the speaker's knowledge. Further, Experiments 2 and 4 supported the second model prediction showing that the words that a speaker tends to use affect the strength of scalar implicature that comprehenders make. In particular, when the speaker has partial knowledge (e.g., has only examined two of three letters), comprehenders think that “some” is more likely to mean “not all” when the speaker also tends to produce number words in similar sentences (e.g., “2 of today's rooms have working smoke detectors.”). These results have important ramifications for theories of meaning: the context beyond the sentence (e.g., the speaker's tendency to use particular words) affects the set of alternatives that comprehenders consider when inferring meaning. (shrink)

What the world needs now is another theory of vagueness. Not because the old theories are useless. Quite the contrary, the old theories provide many of the materials we need to construct the truest theory of vagueness ever seen. The theory shall be similar in motivation to supervaluationism, but more akin to many-valued theories in conceptualisation. What I take from the many-valued theories is the idea that some sentences can be truer than others. But I say very different things to (...) the ordering over sentences this relation generates. I say it is not a linear ordering, so it cannot be represented by the real numbers. I also argue that since there is higher-order vagueness, any mapping between sentences and mathematical objects is bound to be inappropriate. This is no cause for regret; we can say all we want to say by using the comparative truer than without mapping it onto some mathematical objects. From supervaluationism I take the idea that we can keep classical logic without keeping the familiar bivalent semantics for classical logic. But my preservation of classical logic is more comprehensive than is normally permitted by supervaluationism, for I preserve classical inference rules as well as classical sequents. And I do this without relying on the concept of acceptable precisifications as an unexplained explainer. The world does not need another guide to varieties of theories of vagueness, especially since Timothy Williamson (1994) and Rosanna Keefe (2000) have already provided quite good guides. I assume throughout familiarity with popular theories of vagueness. (shrink)

Although the presence or absence of a pitch accent clearly can play an important role in signaling the discourse and information structure of an utterance, whether the form of an accent determines the type of information it conveys is more controversial. We used an eye‐tracking paradigm to investigate whether H*, which has been argued to signal new information, evokes different eye fixations than L+H*, which has been argued to signal the presence of contrast. Our results demonstrate that although listeners interpret (...) these accents differently, their interpretive domains overlap. L+H* creates a strong bias toward contrast referents whereas H* is compatible with both new and contrast referents. (shrink)

This article introduces the topic ‘‘Production of Referring Expressions: Bridging the Gap between Computational and Empirical Approaches to Reference’’ of the journal Topics in Cognitive Science. We argue that computational and psycholinguistic approaches to reference production can benefit from closer interaction, and that this is likely to result in the construction of algorithms that differ markedly from the ones currently known in the computational literature. We focus particularly on determinism, the feature of existing algorithms that is perhaps most clearly at (...) odds with psycholinguistic results, discussing how future algorithms might include non-determinism, and how new psycholinguistic experiments could inform the development of such algorithms. (shrink)

A substantial amount of recent work in natural language generation has focused on the generation of ‘‘one-shot’’ referring expressions whose only aim is to identify a target referent. Dale and Reiter's Incremental Algorithm (IA) is often thought to be the best algorithm for maximizing the similarity to referring expressions produced by people. We test this hypothesis by eliciting referring expressions from human subjects and computing the similarity between the expressions elicited and the ones generated by algorithms. It turns out that (...) the success of the IA depends substantially on the ‘‘preference order’’ (PO) employed by the IA, particularly in complex domains. While some POs cause the IA to produce referring expressions that are very similar to expressions produced by human subjects, others cause the IA to perform worse than its main competitors; moreover, it turns out to be difficult to predict the success of a PO on the basis of existing psycholinguistic findings or frequencies in corpora. We also examine the computational complexity of the algorithms in question and argue that there are no compelling reasons for preferring the IA over some of its main competitors on these grounds. We conclude that future research on the generation of referring expressions should explore alternatives to the IA, focusing on algorithms, inspired by the Greedy Algorithm, which do not work with a fixed PO. (shrink)

Next SectionIn recent years, the interpretation of scalar terms in embedded environments has been investigated extensively. Some experimentalists have been concerned with sentences like (1), in which a scalar term is embedded under a universal quantifier. The controversy involves the question whether ‘some’ in these sentences is interpreted as ‘some but not all’, thus leading to the embedded upper-bounded inference that no square is connected to all of the circles. (1) All the squares are connected with some of the circles.Geurts (...) & Pouscoulous (2009) conducted a verification task that seemed to prove that the inference is not licensed. In response, Clifton & Dube (2010) and Chemla & Spector (2011) gathered evidence suggesting the opposite conclusion. By experimentally investigating the typicality structure of complex quantified sentences like (1) above, I show how the results of the latter two experiments can be explained as typicality effects. Typicality plays a significant role in the comprehension of quantified sentences, thus complicating the interpretation of data from the kinds of experiments that have been employed to test between theories of upper-bounded inferences. (shrink)

In referential communication, Grice's Maxim of Quantity is thought to imply that utterances conveying unnecessary information should incur comprehension difficulties. There is, however, considerable evidence that speakers frequently encode redundant information in their referring expressions, raising the question as to whether such overspecifications hinder listeners’ processing. Evidence from previous work is inconclusive, and mostly comes from offline studies. In this article, we present two event‐related potential (ERP) experiments, investigating the real‐time comprehension of referring expressions that contain redundant adjectives in complex (...) visual contexts. Our findings provide support for both Gricean and bounded‐rational accounts. We argue that these seemingly incompatible results can be reconciled if common ground is taken into account. We propose a bounded‐rational account of overspecification, according to which even redundant words can be beneficial to comprehension to the extent that they facilitate the reduction of listeners’ uncertainty regarding the target referent. (shrink)

A number of authors have defended permissivism by appealing to rational supererogation, the thought that some doxastic states might be rationally permissible even though there are other, more rational beliefs available. If this is correct, then there are situations that allow for multiple rational doxastic responses, even if some of those responses are rationally suboptimal. In this paper, I will argue that this is the wrong approach to defending permissivism—there are no doxastic states that are rationally supererogatory. By the lights (...) of contemporary linguistics, ‘rational’ is an absolute gradable adjective, and as such, can only be applied to things that satisfy the top of the scale of rationality. For this reason, it is not possible to believe what is rational while also failing to believe what is rationally optimal. (shrink)

Formal epistemologists often claim that our credences should be representable by a probability function. Complete probabilistic coherence, however, is only possible for ideal agents, raising the question of how this requirement relates to our everyday judgments concerning rationality. One possible answer is that being rational is a contextual matter, that the standards for rationality change along with the situation. Just like who counts as tall changes depending on whether we are considering toddlers or basketball players, perhaps what counts as rational (...) shifts according to whether we are considering ideal agents or creatures more like ourselves. Even though a number of formal epistemologists have endorsed this type of solution, I will argue that there is no way to spell out this contextual account that can make sense of our everyday judgments about rationality. Those who defend probabilistic coherence requirements will need an alternative account of the relationship between real and ideal rationality. (shrink)

It is clear that beliefs can be assessed both as to their justification and their rationality. What is not as clear, however, is how the rationality and justification of belief relate to one another. Stewart Cohen has stumped for the popular proposal that rationality and justification come to the same thing, that rational beliefs just are justified beliefs, supporting his view by arguing that ‘justified belief’ and ‘rational belief’ are synonymous. In this paper, I will give reason to think that (...) Cohen’s argument is spurious. I will show that ‘rational’ and ‘justified’ occupy two distinct semantic categories – ‘rational’ is an absolute gradable adjective and ‘justified’ is a relative gradable adjective – telling against the thought that ‘rational belief’ and ‘justified belief’ are synonymous. I will then argue that the burden of proof is on those who would equate rationality and justification, making the case that those who hold this prominent position face the difficulty of explaining how rationality and justification come to the same thing even though ‘rational’ and ‘justified’ are not synonymous. (shrink)

Grice's notion of conversational implicature requires that speaker meaning be calculable on the basis of sentence meaning, and presumptions about the speaker's adherence to cooperative principles of conversation and the ability of the hearer to work out the speaker's meaning. However, the actual real‐time consideration of cooperative principles by both the hearer and speaker runs up against severe temporal constraints during language processing. This article considers the role of language processing research in the shaping of a theory of implicature, and (...) provides an empirical overview of pertinent current work in real‐time language production and comprehension. (shrink)

The interplay of content and context is observable in a moment to moment manner as propositional content unfolds. The current contribution illustrates this through data from real-time language comprehension indicating that propositional content is not computed in isolation but relies in important ways on context during every step of the computation of meaning. The relevant notion of context that we have to adopt includes all aspects of possible worlds and draws on a variety of knowledge representations, which in a first (...) processing phase serve to generate expectations for upcoming words. In a second phase, the discourse representation is assessed and if necessary updated by means of inferential reasoning and enrichment to reflect the speaker’s intended meaning. (shrink)

Cognitive Science, Volume 46, Issue 4, April 2022.

Upon hearing a scalar adjective in a definite referring expression such as “the big…,” listeners typically make anticipatory eye movements to an item in a contrast set, such as a big glass in the context of a smaller glass. Recent studies have suggested that this rapid, contrastive interpretation of scalar adjectives is malleable and calibrated to the speaker's pragmatic competence. In a series of eye‐tracking experiments, we explore the nature of the evidence necessary for the modulation of pragmatic inferences in (...) language comprehension, focusing on the complementary roles of top–down information ‐ (knowledge about the particular speaker's pragmatic competence) and bottom‐up cues (distributional information about the use of scalar adjectives in the environment). We find that bottom‐up evidence alone (e.g., the speaker says “the big dog” in a context with one dog), in large quantities, can be sufficient to trigger modulation of the listener's contrastive inferences, with or without top‐down cues to support this adaptation. Further, these findings suggest that listeners track and flexibly combine multiple sources of information in service of efficient pragmatic communication. (shrink)

We investigated how two cues to contrast—beat gesture and contrastive pitch accenting—affect comprehenders' cognitive load during processing of spoken referring expressions. In two visual‐world experiments, we orthogonally manipulated the presence of these cues and their felicity, or fit, with the local (sentence‐level) referential context in critical referring expressions while comprehenders' task‐evoked pupillary responses (TEPRs) were examined. In Experiment 1, beat gesture and contrastive accenting always matched the referential context of filler referring expressions and were therefore relatively felicitous on the global (...) (experiment) level, whereas in Experiment 2, beat gesture and contrastive accenting never fit the referential context of filler referring expressions and were therefore infelicitous on the global level. The results revealed that both beat gesture and contrastive accenting increased comprehenders' cognitive load. For beat gesture, this increase in cognitive load was driven by both local and global infelicity. For contrastive accenting, this increase in cognitive load was unaffected when cues were globally felicitous but exacerbated when cues were globally infelicitous. Together, these results suggest that comprehenders' cognitive resources are taxed by processing infelicitous use of beat gesture and contrastive accenting to convey contrast on both the local and global levels. (shrink)

Evidence from numerous studies using the visual world paradigm has revealed both that spoken language can rapidly guide attention in a related visual scene and that scene information can immediately influence comprehension processes. These findings motivated the coordinated interplay account (Knoeferle & Crocker, 2006) of situated comprehension, which claims that utterance‐mediated attention crucially underlies this closely coordinated interaction of language and scene processing. We present a recurrent sigma‐pi neural network that models the rapid use of scene information, exploiting an utterance‐mediated (...) attentional mechanism that directly instantiates the CIA. The model is shown to achieve high levels of performance (both with and without scene contexts), while also exhibiting hallmark behaviors of situated comprehension, such as incremental processing, anticipation of appropriate role fillers, as well as the immediate use, and priority, of depicted event information through the coordinated use of utterance‐mediated attention to the scene. (shrink)

Cognitive Science, Volume 46, Issue 1, January 2022.

Beyond the observation that both speakers and listeners rapidly inspect the visual targets of referring expressions, it has been argued that such gaze may constitute part of the communicative signal. In this study, we investigate whether a speaker may, in principle, exploit listener gaze to improve communicative success. In the context of a virtual environment where listeners follow computer-generated instructions, we provide two kinds of support for this claim. First, we show that listener gaze provides a reliable real-time index of (...) understanding even in dynamic and complex environments, and on a per-utterance basis. Second, we show that a language generation system that uses listener gaze to provide rapid feedback improves overall task performance in comparison with two systems that do not use gaze. Aside from demonstrating the utility of listener gaze in situated communication, our findings open the door to new methods for developing and evaluating multi-modal models of situated interaction. (shrink)

Speakers often overspecify by encoding more information than is necessary when referring to an object (e.g., “the blue mug” for the only mug in a group of objects). We investigated the role of a partner's linguistic behavior (whether or not they overspecify) on a speaker's own tendency to overspecify. We used a director–matcher task in which speakers interacted with a partner who either consistently overspecified or minimally specified in the color/size dimension (Experiments 1, 2, and 3), as well as with (...) a partner who switched behaviors midway through interaction (Experiments 4 and 5). We found that speakers aligned with their partner's linguistic behavior to produce overspecific or minimally specific descriptions, and we saw little evidence that the alignment was enhanced by lexical or semantic repetition across prime and target trials. Time‐course analyses showed that alignment increased over the course of the interaction, and speakers appeared to track a change in the partner's linguistic behavior, altering their reference strategy to continue matching that of their partner's. These results demonstrate the persistent influence of a partner's behavior on speakers across the duration of an interaction. (shrink)

A central goal of typological research is to characterize linguistic features in terms of both their functional role and their fit to social and cognitive systems. One long-standing puzzle concerns why certain languages employ grammatical gender. In an information theoretic analysis of German noun classification, Dye, Milin, Futrell, and Ramscar enumerated a number of important processing advantages gender confers. Yet this raises a further puzzle: If gender systems are so beneficial to processing, what does this mean for languages that make (...) do without them? Here, we compare the communicative function of gender marking in German to that of prenominal adjectives in English, finding that despite their differences, both systems act to efficiently smooth information over discourse, making nouns more equally predictable in context. We examine why evolutionary pressures may favor one system over another and discuss the implications for compositional accounts of meaning and Gricean principles of communication. (shrink)

Semantic memory encompasses one's knowledge about the world. Distributional semantic models, which construct vector spaces with embedded words, are a proposed framework for understanding the representational structure of human semantic knowledge. Unlike some classic semantic models, distributional semantic models lack a mechanism for specifying the properties of concepts, which raises questions regarding their utility for a general theory of semantic knowledge. Here, we develop a computational model of a binary semantic classification task, in which participants judged target words for the (...) referent's size or animacy. We created a family of models, evaluating multiple distributional semantic models, and mechanisms for performing the classification. The most successful model constructed two composite representations for each extreme of the decision axis (e.g., one averaging together representations of characteristically big things and another of characteristically small things). Next, the target item was compared to each composite representation, allowing the model to classify more than 1,500 words with human‐range performance and to predict response times. We propose that when making a decision on a binary semantic classification task, humans use task prompts to retrieve instances representative of the extremes on that semantic dimension and compare the probe to those instances. This proposal is consistent with the principles of the instance theory of semantic memory. (shrink)

Three experiments investigated the processing of the implicature associated with some using a “gumball paradigm.” On each trial, participants saw an image of a gumball machine with an upper chamber with 13 gumballs and an empty lower chamber. Gumballs then dropped to the lower chamber and participants evaluated statements, such as “You got some of the gumballs.” Experiment 1 established that some is less natural for reference to small sets and unpartitioned sets compared to intermediate sets. Partitive some of was (...) less natural than simple some when used with the unpartitioned set. In Experiment 2, including exact number descriptions lowered naturalness ratings for some with small sets but not for intermediate size sets and the unpartitioned set. In Experiment 3, the naturalness ratings from Experiment 2 predicted response times. The results are interpreted as evidence for a Constraint-Based account of scalar implicature processing and against both two-stage, Literal-First models and pragmatic Default models. (shrink)

A substantial amount of recent work in natural language generation has focused on the generation of ‘‘one‐shot’’ referring expressions whose only aim is to identify a target referent. Dale and Reiter's Incremental Algorithm (IA) is often thought to be the best algorithm for maximizing the similarity to referring expressions produced by people. We test this hypothesis by eliciting referring expressions from human subjects and computing the similarity between the expressions elicited and the ones generated by algorithms. It turns out that (...) the success of the IA depends substantially on the ‘‘preference order’’ (PO) employed by the IA, particularly in complex domains. While some POs cause the IA to produce referring expressions that are very similar to expressions produced by human subjects, others cause the IA to perform worse than its main competitors; moreover, it turns out to be difficult to predict the success of a PO on the basis of existing psycholinguistic findings or frequencies in corpora. We also examine the computational complexity of the algorithms in question and argue that there are no compelling reasons for preferring the IA over some of its main competitors on these grounds. We conclude that future research on the generation of referring expressions should explore alternatives to the IA, focusing on algorithms, inspired by the Greedy Algorithm, which do not work with a fixed PO. (shrink)