Canada

A central question in cognitive science is how semantic information is mentally represented. Two dominant theories of semantic representation are language-based distributional semantic models (which suggest that word meaning is based on which words co-occur in language) and semantic networks based on word associations (which suggest that words are represented as a network in which words with closer meanings are more closely linked). We investigate the issue of semantic representation through the lens of core vocabulary -- the set of words that are most central in the mental lexicon -- which these two theories make different predictions about. We report on the results of an experiment that tests which measure of core vocabulary most closely aligns with human behaviour in a word-guessing game where the aim was to identify a target word given a set of semantically related words as hints. Target and hint words, which varied across trials, were generated from different core vocabulary lists corresponding to these different theories. Results revealed that the type of hint words did not affect performance, but that better performance was attained for target words derived from word associations than from natural language distributional statistics. Follow-up analyses ruled out several alternate explanations. Our results suggest that the semantic information reflected in word associations may be more involved in the efficient identification of lexical meaning.

CogSci 2022

Core words in semantic representation

technical paper

Join us for CogSci2022! Our 44th Annual Conference will be centered around the theme of diversity, with an accompanying range of keynotes and invited speakers who themselves highlight the diversity of our field. Our three symposia are Cognition across Diverse Populations, Cognition within Diverse Societies, and Global Challenges for Cognitive Science. The conference is being offered in a fully hybrid environment for onsite and online attendees.
 
 [**The Program at a Glance can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2022/07/Cogsci-2022-PAAG-Detailed-July-12final-1.pdf)
 [**The Detailed Program can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2022/08/DeArmond-CogSci2022-program-v113-final.pdf)

This conference requires a registration to attend. Please use the link below to secure your ticket. The website will be available to all registered attendees on July 25th. 

[Register to Attend](https://cognitivesciencesociety.org/registration/)

Registration Is Required

The 44th Annual Meeting of the Cognitive Science Society highlighted research on the theme Cognitive Diversity in addition to the full breadth and diversity of research topics offered by the society’s membership. 





To enable natural and fluid human-robot interactions, robots need to not only be able to communicate with humans through natural language, but also do so in a way that complies with the norms of human interaction, such as politeness norms. Doing so is particularly challenging, however, in part due to the sensitivity of such norms to a host of different contextual and intentional factors. In this work, we explore computational models of context-sensitive human politeness norms, using explainable machine learning models to demonstrate the value of both speaker intention and task context in predicting adherence with indirect speech norms. We argue that this type of model, if integrated into a robot cognitive architecture, could be highly successful at enabling robots to predict when they themselves should similarly adhere to these norms.

Leveraging Intentional Factors and Task Context to Predict Linguistic Norm Adherence

Can we know a word by the company it keeps? Aspects of meaning that concern physical interactions might be particularly difficult to learn from language alone. Glenberg & Robertson (2000) found that although human comprehenders were sensitive to the distinction between afforded and nonafforded actions, distributional semantic models were not. We tested whether technological advances have made distributional models more sensitive to affordances by replicating their experiment with modern Neural Language Models (NLMs). We found that only one NLM (GPT-3) was sensitive to the affordedness of actions. Moreover, GPT-3 accounted for only one third of the effect of affordedness on human sensibility judgments. These results imply that people use processes that go beyond distributional statistics to understand linguistic expressions, and that NLP systems may need to be augmented with such capabilities.

Distrubutional Semantics Still Can't Account for Affordances

Do people attribute functions to gendered social categories? (For instance, is there something men or mothers are for?) And if so, do such attributions of function have consequences for normative judgments about what members of these social categories ought to do? In the current study, participants (N = 366) rated their agreement with 15 statements about the “true functions” of different social categories, in triads of matched masculine, feminine, and superordinate categories (e.g., fathers, mothers, and parents). Participants endorsed functional claims more for some social categories (e.g., parents) than others (e.g., kids), and their background beliefs about gender predicted variation in functional reasoning. However, across categories, participants judged that fulfilling true functions was ‘natural’ for members of the category, and they judged that category members ought to fulfill their true functions.

What Are Men and Mothers For? The Causes and Consequences of Functional Reasoning About Social Categories

Word meanings extend over time due to a functional need for maintaining communicative expressivity within a compact lexicon. Previous scholars have suggested that word meanings extend via a process of chaining, whereby novel items link to existing ones close in semantic space. Recent work has formalized this idea using computational models grounded typically in the exemplar and prototype theories of categorization
that are either memory-intensive or simplistic in representation. We propose an alternative account of chaining that optimizes cognitive efficiency by trading off representational accuracy with memory complexity. We operationalize this efficient chaining as an infinite mixture model and show how it constructs the internal representations of word meaning adaptively through time while predicting the historical development of English verb meanings with precision and limited resources.

Infinite mixture chaining: Efficient temporal construction of word meaning

Artificial language learning research has shown that, under some conditions, adult speakers tend to probability-match to inconsistent variation in their input, while in others, they regularize by reducing that variation. We demonstrate that this framework can characterize speaker behavior in a natural-language morphological inflection task: 
the lexicon can be used to estimate variation in speaker productions. In the task of German plural inflection, we find that speakers probability-match a lexical distribution conditioned on phonology, and largely disregard an alternative possible strategy of conditional regularization based on grammatical gender.

Regularization or lexical probability-matching? How German speakers generalize plural morphology

Prior work has shown a “natural” preference in the Verb Phrase for direct object Nouns to linearly precede the Verb. There is also evidence of a “natural” preference in the Noun Phrase to order Nouns before Adjectives. Given this, we asked how domain-general biases like regularization and language-specific biases like the preference for “natural” orders could jointly contribute to the emergence of these two common word orders cross-linguistically. Using a silent gesture paradigm (in which we presented iconic gestures without speech), we exposed different participants to competing Verb Phrase (NounVerb vs. VerbNoun) and Noun Phrase (NounAdj vs. AdjNoun) word orders at varying frequencies. In Noun Phrase contrast conditions, we found that regularization was greatest when the domain-general bias towards regularization and the linguistic bias to order Nouns before Adjectives were aligned. In Verb Phrase conditions, participants regularized to the same extent regardless of input: They opted for greater regularity, even at the expense of aligning with underlying word order biases. We discuss the implications of our work for understanding the effects of domain-general biases on language.

Regularization of Word Order in the Verb Phrase differs from the Noun Phrase: Evidence from an online silent gesture perception paradigm

Languages often express grammatical information through inflectional morphology, in which grammatical features are grouped into strings of morphemes. In this work, we propose that cross-linguistic generalizations about morphological fusion, in which multiple features are expressed through one morpheme, can be explained in part by optimization of processing efficiency, as formalized using the memory--surprisal tradeoff of Hahn et al. (2021). We show in a toy setting that fusion of highly informative neighboring morphemes can lead to greater processing efficiency under our processing model. Next, based on paradigm and frequency data from four languages, we consider both total fusion and gradable fusion using empirical measures developed by Rathi et al. (2021), and find that the degree of fusion is predicted by closeness of optimal morpheme ordering as determined by optimization of processing efficiency. Finally, we show that optimization of processing efficiency can successfully predict typological patterns involving suppletion.

Explaining patterns of fusion in morphological paradigms using the memory--surprisal tradeoff

Responsive predicates are clause-embedding predicates like English 'know' and 'guess' that can take both declarative and interrogative clausal complements. The meanings of responsive predicates when they take a declarative complement and when they take an interrogative complement are hypothesized to be constrained in systematic ways across languages, suggesting that these constraints represent semantic universals. We report an artificial language learning experiment showing that one of these proposed constraints is indeed reflected in the inferences participants make while learning a novel responsive predicate. Our results add support to a growing body of evidence linking semantic universals to learning.

Learnability and constraints on the semantics of clause-embedding predicates

Rumelhart Symposium

Humans are surprisingly adept at interpretingwhat is happening around them –theyspontaneously and rapidlysegmentand organizetheirdynamic experience into coherent event construals. Such event construals may offer a starting point for assembling a linguistic description of the event during speaking (Levelt, 1989). However, the precise format of event representations and their mapping to language have remained elusive, partly because research on how people mentally segment and perceive events (see Radvansky & Zacks, 2014 for a review) has largely proceeded separately from analyses of how events are encoded in language (see Truswell, 2019 for a review).

Premium content

Core words in semantic representation

Downloads

Next from CogSci 2022

Leveraging Intentional Factors and Task Context to Predict Linguistic Norm Adherence

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES