Search

This paper presents a new cross-linguistic generalization in the domain of modality and explores its explanation in terms of communicative efficiency. Based on original data elicitation, we observe that, if a language lexicalizes any impossibilities, then it lexicalizes deontic impossibility. Our computational modeling suggests that this priority for deontic flavor is accounted for under the communicative efficiency model if a bias for deontic flavor is included in terms of utility values. Our results contribute to the empirical landscape of modal semantic typology and the growing literature on the role of bias in the communicative efficiency analysis for cross-linguistic lexicalization patterns

A semantic universal, which we here dub the Veridical Uniformity Universal, has recently been argued to hold of responsive verbs (those that take both declarative and interrogative complements). This paper offers a preliminary explanation of this universal: verbs satisfying it are easier to learn than those that do not. This claim is supported by a computational experiment using artificial neural networks, mirroring a recent proposal for explaining semantic universals of quantifiers. This preliminary study opens up many avenues for future work on explaining semantic universals more generally, which are discussed in the conclusion.

The vocabulary of human languages has been argued to support efficient communication by optimizing the trade-off between simplicity and informativeness. The argument has been originally based on cross-linguistic analyses of vocabulary in semantic domains of content words, such as kinship, color, and number terms. The present work applies this analysis to a category of function words: indefinite pronouns (e.g., someone, anyone, no one). We build on previous work to establish the meaning space and featural make-up for indefinite pronouns, and show that indefinite pronoun systems across languages optimize the simplicity/informativeness trade-off. This demonstrates that pressures for efficient communication shape both content and function word categories. In doing so, our work aligns with several concurrent studies exploring the simplicity/informativeness trade-off in functional vocabulary. Importantly, we further argue that the trade-off may explain some of the universal properties of indefinite pronouns, thus reducing the explanatory load for linguistic theories.

The meanings expressed by the world’s languages have been argued to support efficient communication. Evidence for this hypothesis has drawn on cross-linguistic analyses of vocabulary in semantic domains of both content words (e.g. kinship terms (Kemp & Regier 2012); color terms (Regier, Kay & Khetarpal 2007; Zaslavsky, Kemp, Regier & Tishby 2018)) and function words (e.g.quantifiers(Steinert-Threlkeld2021); indefinite pronouns(Deni ́c, Steinert-Threlkeld & Szymanik 2022)) approaching the hypothesis concretely in terms of a trade-off between simplicity and informativeness. We apply the analysis to modals (e.g. can, ought, might). Two proposed universals in this domain from Nauze (2008) and Vander Klok (2013) are used for generating many artificial languages with varying degrees of quasi-naturalness as a proxy for natural data. A computational experiment shows that most of the optimal solutions to the trade-off problem are predicted by Vander Klok; meanwhile, as languages more robustly satisfy Nauze’s universal, they also become more optimal. This suggests that efficient communication is a leading explanation for constraints on modal semantic variation.

Natural languages exhibit many semantic universals, that is, properties of meaning shared across all languages. In this paper, we develop an explanation of one very prominent semantic universal, the monotonicity universal. While the existing work has shown that quantifiers satisfying the monotonicity universal are easier to learn, we provide a more complete explanation by considering the emergence of quantifiers from the perspective of cultural evolution. In particular, we show that quantifiers satisfy the monotonicity universal evolve reliably in an iterated learning paradigm with neural networks as agents.

Expressivists about epistemic modals deny that ‘Jane might be late’ canonically serves to express the speaker’s acceptance of a certain propositional content. Instead, they hold that it expresses a lack of acceptance (that Jane isn’t late). Prominent expressivists embrace pragmatic expressivism: the doxastic property expressed by a declarative is not helpfully identified with (any part of) that sentence’s compositional semantic value. Against this, we defend semantic expressivism about epistemic modals: the semantic value of a declarative from this domain is (partly) the property of doxastic attitudes it canonically serves to express. In support, we synthesize data from the critical literature on expressivism—largely reflecting interactions between modals and disjunctions—and present a semantic expressivism that readily predicts the data. This contrasts with salient competitors, including: pragmatic expressivism based on domain semantics or dynamic semantics; semantic expressivism à la Moss (Semant Pragmat 8(5):1–81, 2015. 10.3765/sp.8.5); and the bounded relational semantics of Mandelkern (Philos Rev 128(1):1–61, 2019. 10.1215/00318108-7213001).

Author(s): van de Pol, Iris; Lodder, Paul; van Maanen, Leendert; Steinert-Threlkeld, Shane; Szymanik, Jakub | Abstract: Despite wide variation among natural languages, there are linguistic properties thought to be universal to all or almost all natural languages. Here, we consider universals at the semantic level, in the domain of quantifiers, which are given by the properties of monotonicity, quantity, and conservativity. We investigate whether these universals might be explained by differences in complexity. We generate a large collection of quantifiers, based on a simple yet expressive grammar, and compute both their complexities and whether they adhere to these universal properties. We find that quantifiers satisfying semantic universals are less complex: they have a shorter minimal description length.

This work presents methods for learning cross-lingual sentence representations using paired or unpaired bilingual texts. We hypothesize that the cross-lingual alignment strategy is transferable, and therefore a model trained to align only two languages can encode multilingually more aligned representations. We thus introduce dual-pivot transfer: training on one language pair and evaluating on other pairs. To study this theory, we design unsupervised models trained on unpaired sentences and single-pair supervised models trained on bitexts, both based on the unsupervised language model XLM-R with its parameters frozen. The experiments evaluate the models as universal sentence encoders on the task of unsupervised bitext mining on two datasets, where the unsupervised model reaches the state of the art of unsupervised retrieval, and the alternative single-pair supervised model approaches the performance of multilingually supervised models. The results suggest that bilingual training techniques as proposed can be applied to get sentence representations with multilingual alignment., Comment: To be published at ACL 2022. 11 pages, 2 figures

International audience; In recent years, the methods of formal semantics and pragmatics have been fruitfully applied to the analysis of primate communication systems. Most analyses therein appeal to a division of labor between semantics and pragmatics which has the following three features: (F1) calls are given referential meanings (they provide information about the world rather than just about an action to be taken), (F2) some calls have a general meaning, and (F3) the meanings of calls in context are enriched by competition with more informative calls, along the lines of scalar implicatures. In this paper, we develop highly simplified models to independently assess the conditions under which such features would emerge. After identifying a sufficient condition for (F1), we find a range of conditions under which (F2) and (F3) are not evolutionarily stable, and discuss the consequences for both modeling and empirical work.

We propose a general framework to study language emergence through signaling games with neural agents. Using a continuous latent space, we are able to (i) train using backpropagation, (ii) show that discrete messages nonetheless naturally emerge. We explore whether categorical perception effects follow and show that the messages are not compositional., Comment: ACL 2020

Although large-scale pretrained language models, such as BERT and RoBERTa, have achieved superhuman performance on in-distribution test sets, their performance suffers on out-of-distribution test sets (e.g., on contrast sets). Building contrast sets often re-quires human-expert annotation, which is expensive and hard to create on a large scale. In this work, we propose a Linguistically-Informed Transformation (LIT) method to automatically generate contrast sets, which enables practitioners to explore linguistic phenomena of interests as well as compose different phenomena. Experimenting with our method on SNLI and MNLI shows that current pretrained language models, although being claimed to contain sufficient linguistic knowledge, struggle on our automatically generated contrast sets. Furthermore, we improve models' performance on the contrast sets by apply-ing LIT to augment the training data, without affecting performance on the original data., Comment: Appears at EMNLP BlackboxNLP Workshop 2020

Despite wide variation among natural languages, there are linguistic properties universal to all (or nearly all) languages. An important challenge is to explain why these linguistic universals hold. One explanation employs a learnability argument: semantic universals hold because expressions that satisfy them are easier to learn than those that do not. In an exploratory study we investigate the relation between learnability and complexity and whether the presence of semantic universals for quantifiers can also be explained by differences in complexity. We develop a novel application of (approximate) Kolmogorov complexity to measure fine-grained distinctions in complexity between different quantifiers. Our results indicate that the monotonicity universal can be explained by complexity while the conservativity universal cannot. For quantity we did not find a robust result. We also found that learnability and complexity pattern together in the monotonicity and conservativity cases that we consider, while that pattern is less robust in the quantity cases.

How are the meanings of linguistic expressions related to their use in concrete cognitive tasks? Visual identification tasks show human speakers can exhibit considerable variation in their understanding, representation and verification of certain quantifiers. This paper initiates an investigation into neural models of these psycho-semantic tasks. We trained two types of network – a convolutional neural network (CNN) model and a recurrent model of visual attention (RAM) – on the “most” verification task from Pietroski2009, manipulating the visual scene and novel notions of task duration. Our results qualitatively mirror certain features of human performance (such as sensitivity to the ratio of set sizes, indicating a reliance on approximate number) while differing in interesting ways (such as exhibiting a subtly different pattern for the effect of image type). We conclude by discussing the prospects for using neural models as cognitive models of this and other psychosemantic tasks.

One of the great successes of the application of generalized quantifiers to natural language has been the ability to formulate robust semantic universals. When such a universal is attested, the question arises as to the source of the universal. In this paper, we explore the hypothesis that many semantic universals arise because expressions satisfying the universal are easier to learn than those that do not. While the idea that learnability explains universals is not new, explicit accounts of learning that can make good on this hypothesis are few and far between. We propose a model of learning — back-propagation through a recurrent neural network — which can make good on this promise. In particular, we discuss the universals of monotonicity, quantity, and conservativity and perform computational experiments of training such a network to learn to verify quantifiers. Our results are able to explain monotonicity and quantity quite well. We suggest that conservativity may have a different source than the other universals. EARLY ACCESS

The vocabulary of human languages has been argued to support efficient communication by optimizing the trade-off between complexity and informativeness (Kemp & Regier 2012). The argument has been based on cross-linguistic analyses of vocabulary in semantic domains of content words such as kinship, color, and number terms. The present work extends this analysis to a category of function words: indefinite pronouns (e.g. someone, anyone, no-one, cf. Haspelmath 2001). We build on previous work to establish the meaning space and featural make-up for indefinite pronouns, and show that indefinite pronoun systems across languages optimize the complexity/informativeness trade-off. This demonstrates that pressures for efficient communication shape both content and function word categories, thus tying in with the conclusions of recent work on quantifiers by Steinert-Threlkeld (2019). Furthermore, we argue that the trade-off may explain some of the universal properties of indefinite pronouns, thus reducing the explanatory load for linguistic theories.

We investigate, in a logical setting, the expressivist proposal that assertion primarily functions to express and coordinate doxastic states and that ‘might’ fundamentally expresses lack of belief. We provide a formal model of an agent’s doxastic state and novel assertability conditions for an associated formal language. We thereby prove that an arbitrary assertion (including a complex of ‘might’ and ‘factual’ claims) always succeeds in expressing a well-defined (partial) doxastic state, and propose a fully general and intuitive update operation as a model of an agent coming to accept an arbitrary assertion. Leaving a comprehensive philosophical and linguistic defense for elsewhere, we explore technical aspects of our framework, providing, for instance, a complete logic of assertability and reduction axioms for the novel update operations related to our proposal. Finally, we contrast our work with related proposals in the logic literature.

A special kind of substitution on languages called iteration is presented and studied. These languages arise in the application of semantic automata to iterations of generalized quantifiers. We show that each of the star-free, regular, and deterministic context-free languages are closed under iteration and that it is decidable whether a given regular or determinstic context-free language is an iteration of two such languages. This result can be read as saying that the van Benthem/Keenan `Frege Boundary' is decidable for large subclasses of natural language quantifiers. We also determine the state complexity of iteration of regular languages.

Semantic universals are properties of meaning shared by the languages of the world. We offer an explanation of the presence of such universals by measuring simplicity in terms of ease of learning, showing that expressions satisfying universals are simpler than those that do not according to this criterion. We measure ease of learning using tools from machine learning and analyze universals in a domain of function words (quantifiers) and content words (color terms). Our results provide strong evidence that semantic universals across both function and content words reflect simplicity as measured by ease of learning.

The semantic automata framework, developed originally in the 1980s, provides computational interpretations of generalized quantifiers. While recent experimental results have associated structural features of these automata with neuroanatomical demands in processing sentences with quantifiers, the theoretical framework has remained largely unexplored. In this paper, after presenting some classic results on semantic automata in a modern style, we present the first application of semantic automata to polyadic quantification, exhibiting automata for iterated quantifiers. We also discuss the role of semantic automata in linguistic theory and offer new empirical predictions for sentence processing with embedded quantifiers.

We investigate, in a logical setting, the proposal that assertion primarily functions to express and coordinate doxastic states and that ‘might’ fundamentally expresses lack of belief. We provide a formal model of an agent’s doxastic state and precise assertability conditions for an associated formal language. We thereby prove that an arbitrary assertion (including a complex of ‘might’ and ‘factual’ claims) always succeeds in expressing a well-defined doxastic state. We then propose a fully general and intuitive doxastic update operation as a model of an agent coming to accept an arbitrary assertion. We provide reduction axioms for some novel update operations related to this proposal.

Graphical abstractDisplay Omitted Highlights? Visual inspection found significant expansion in mid-anterior of left ventricle. ? High resolution human LV atlas labeled with terms from large reference ontology. ? Atlas mapped to previous study's coordinates. Semantic tools used to query. ? Automated methods found same region in apical anterior. Other metrics agree. ? A formal, precise, automated, and reproducible way to make location-based inferences. A method for automated location of shape differences in diseased anatomical structures via high resolution biomedical atlases annotated with labels from formal ontologies is described. In particular, a high resolution magnetic resonance image of the myocardium of the human left ventricle was segmented and annotated with structural terms from an extracted subset of the Foundational Model of Anatomy ontology. The atlas was registered to the end systole template of a previous study of left ventricular remodeling in cardiomyopathy using a diffeomorphic registration algorithm. The previous study used thresholding and visual inspection to locate a region of statistical significance which distinguished patients with ischemic cardiomyopathy from those with nonischemic cardiomyopathy. Using semantic technologies and the deformed annotated atlas, this location was more precisely found. Although this study used only a cardiac atlas, it provides a proof-of-concept that ontologically labeled biomedical atlases of any anatomical structure can be used to automate location-based inferences.

Shane Steinert-Threlkeld∗, Siamak Ardekani∗, Jose L.V. Mejino‡, Landon T. Detwiler‡, James F. Brinkley‡, Michael Halle¶, Ron Kikinis¶, Raimond L. Winslow†, Michael I. Miller∗, J. Tilak Ratnanather∗ ∗Center for Imaging Science, Johns Hopkins University, Baltimore, MD †Institute for Computational Medicine, Johns Hopkins University, Baltimore, MD ‡Structural Informatics Group, University of Washington, Seattle, WA §Department of Biological Structure, University of Washington, Seattle, WA ¶Surgical Planning Laboratory, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA

All natural languages exhibit a distinction between content words (nouns, verbs, etc.) and function words (determiners, auxiliaries, tenses, etc.). Yet surprisingly little has been said about the emergence of this universal architectural feature of human language. This article argues that the existence of this distinction requires the presence of nontrivial compositionality and identifies assumptions that have previously been made in the literature that provably guarantee only trivial composition. It then presents a signaling game with variable contexts and shows how the distinction can emerge via reinforcement learning.