Jennifer Culbertson
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Jennifer Culbertson has a total of 39 co-authors. Here are the top co-authors by number of shared publications:
- Kenny Smith
The University of Edinburgh
7 shared publications - Alexander Martin
Rijksuniversiteit Groningen
4 shared publications Géraldine Legendre
4 shared publicationsPaul Smolensky
4 shared publications- Simon Kirby
Centre for Language Evolution, University of Edinburgh, Edinburgh, UK
4 shared publications - Annie Holtz
Department of Linguistics and English Language The University of Edinburgh
3 shared publications
Latest Publications
- Open Access
Agreement Marking May Facilitate Noun Class Learning in Children But Not Adults
Agreement, defined as the systematic covariance between elements (e.g., between the 3rd person singular subject she and the -s verb ending in ‘she writes’), is present in many languages despite introducing redundant complexity. A prominent hypothesis suggests that the ubiquity of complex morphological features such as agreement might be attributable to a functional advantage it provides specifically to child learners (Lupyan & Dale, 2010). We tested this hypothesis through two artificial language learning experiments involving both children and adults. In both studies we compared noun class learning in artificial languages with and without agreement marking. While no overall advantage of agreement in noun classification was observed, our findings indicate that agreement may have a differential effect on children and adults. Specifically, agreement may facilitate the generalization of noun classification to novel nouns in children, but not in adults. These findings add to the growing body of evidence that highlights differences in the cues children and adults rely on in language acquisition. We discuss the implications of these results for theories of language acquisition and language change.
Open AccessAgreement Marking May Facilitate Noun Class Learning in Children But Not Adults
Agreement, defined as the systematic covariance between elements (e.g., between the 3rd person singular subject she and the -s verb ending in ‘she writes’), is present in many languages despite introducing redundant complexity. A prominent hypothesis suggests that the ubiquity of complex morphological features such as agreement might be attributable to a functional advantage it provides specifically to child learners (Lupyan & Dale, 2010). We tested this hypothesis through two artificial language learning experiments involving both children and adults. In both studies we compared noun class learning in artificial languages with and without agreement marking. While no overall advantage of agreement in noun classification was observed, our findings indicate that agreement may have a differential effect on children and adults. Specifically, agreement may facilitate the generalization of noun classification to novel nouns in children, but not in adults. These findings add to the growing body of evidence that highlights differences in the cues children and adults rely on in language acquisition. We discuss the implications of these results for theories of language acquisition and language change.
- 1 July 2026
- Published by Cambridge University Press (CUP) in Behavioral and Brain Sciences
- Vol. 49, e205
Language models do not yet achieve explanatory adequacy because language is more than incremental prediction
In this commentary, we challenge the idea that Language Models (LMs) provide explanatorily adequate models of human language. Findings from language evolution show that both humans and LMs fail to produce human-like language via inductive biases alone; the communicative function of language is crucial. More generally, both experimental and computational work underscore the fact that language is more than just incremental prediction.
- 1 July 2026
- Published by Cambridge University Press (CUP) in Behavioral and Brain Sciences
- Vol. 49, e205
Language models do not yet achieve explanatory adequacy because language is more than incremental prediction
In this commentary, we challenge the idea that Language Models (LMs) provide explanatorily adequate models of human language. Findings from language evolution show that both humans and LMs fail to produce human-like language via inductive biases alone; the communicative function of language is crucial. More generally, both experimental and computational work underscore the fact that language is more than just incremental prediction.
- Open Access
- 27 January 2026
- Published by California Digital Library (CDL) in Glossa Psycholinguistics
- Vol. 5 (1)
Category-specific and system-wide preferences in competition: Evidence from noun phrase harmony
Typological data show a tendency for languages to exhibit harmonic (i.e., consistent) ordering between heads and dependents. Previous experimental work using artificial language learning experiments has shown that learners prefer harmonic patterns. This suggests that the typological trend for harmony may, at least in part, be driven by a cognitive bias. However, it is well-documented that specific categories sometimes contradict this tendency. Here, we investigate one such case in the domain of the noun phrase. While many nominal dependents exhibit harmony, adjectives and genitives do not: adjectives tend to follow the noun and genitives tend to precede. Previous experiments have identified the existence of cognitive biases that keep these dependents split across the head noun in contexts where there is no conventional language system in place. In this study, we use a silent gesture experiment to examine whether the specific biases that apply to these two dependent types compete with a general preference for harmony in an artificial language learning task. Specifically, we examine whether participants’ learning behaviour is consistent with a preference not just for harmony, but for a non-harmonic order where adjectives follow and genitives precede the noun. What we find, instead, is that participants’ preference for consistent language systems is not modulated by category-specific biases for prenominal genitives and postnominal adjectives. We discuss the implications of this finding for explanations of typological tendencies which link them to cognitive biases.
- 27 January 2026
- Published by California Digital Library (CDL) in Glossa Psycholinguistics
- Vol. 5 (1)
Open AccessCategory-specific and system-wide preferences in competition: Evidence from noun phrase harmony
Typological data show a tendency for languages to exhibit harmonic (i.e., consistent) ordering between heads and dependents. Previous experimental work using artificial language learning experiments has shown that learners prefer harmonic patterns. This suggests that the typological trend for harmony may, at least in part, be driven by a cognitive bias. However, it is well-documented that specific categories sometimes contradict this tendency. Here, we investigate one such case in the domain of the noun phrase. While many nominal dependents exhibit harmony, adjectives and genitives do not: adjectives tend to follow the noun and genitives tend to precede. Previous experiments have identified the existence of cognitive biases that keep these dependents split across the head noun in contexts where there is no conventional language system in place. In this study, we use a silent gesture experiment to examine whether the specific biases that apply to these two dependent types compete with a general preference for harmony in an artificial language learning task. Specifically, we examine whether participants’ learning behaviour is consistent with a preference not just for harmony, but for a non-harmonic order where adjectives follow and genitives precede the noun. What we find, instead, is that participants’ preference for consistent language systems is not modulated by category-specific biases for prenominal genitives and postnominal adjectives. We discuss the implications of this finding for explanations of typological tendencies which link them to cognitive biases.
- Open Access
Domain-general categorisation explains constrained cross-linguistic variation in noun classification
- Open Access
- 26 November 2025
- Published by Wiley in Cognitive Science
- Vol. 49 (11) , e70139
With or Without a System: How Category‐Specific and System‐Wide Cognitive Biases Shape Word Order
Certain recurrent features of language characterize the way a whole language system is structured. By contrast, others target specific categories of items within those wider systems. For example, languages tend to exhibit consistent order of heads and dependents across different phrases—a system‐wide regularity known as harmony. While this tendency is generally robust, some specific syntactic categories appear to deviate from the trend. We examine one such case, the order of adjectives and genitives, which do not exhibit a typological tendency for consistent order with respect to the noun. Instead, adjectives tend to follow and genitives precede the noun. Across two silent gesture experiments, we test the hypothesis that these category‐specific ordering tendencies reflect cognitive biases that favor (i) conveying objects before properties that modify them, but (ii) conveying expressions of possession before possessors. While our hypothesis is thus that these biases are semantic in nature—they impact preferences for how concepts are ordered—the claim is that they may have downstream effects on conventionalized syntax by contributing to an over‐representation of postnominal adjectives and prenominal genitives. We find that these biases affect gesture order in contexts where no conventionalized system is in place. When a system is in place, participants learn that system, and category‐specific biases do not impact their learning. Our results suggest that different contexts reveal distinct types of cognitive biases; some are active during learning and others are active during language creation.
- 26 November 2025
- Published by Wiley in Cognitive Science
- Vol. 49 (11) , e70139
Open AccessWith or Without a System: How Category‐Specific and System‐Wide Cognitive Biases Shape Word Order
Certain recurrent features of language characterize the way a whole language system is structured. By contrast, others target specific categories of items within those wider systems. For example, languages tend to exhibit consistent order of heads and dependents across different phrases—a system‐wide regularity known as harmony. While this tendency is generally robust, some specific syntactic categories appear to deviate from the trend. We examine one such case, the order of adjectives and genitives, which do not exhibit a typological tendency for consistent order with respect to the noun. Instead, adjectives tend to follow and genitives precede the noun. Across two silent gesture experiments, we test the hypothesis that these category‐specific ordering tendencies reflect cognitive biases that favor (i) conveying objects before properties that modify them, but (ii) conveying expressions of possession before possessors. While our hypothesis is thus that these biases are semantic in nature—they impact preferences for how concepts are ordered—the claim is that they may have downstream effects on conventionalized syntax by contributing to an over‐representation of postnominal adjectives and prenominal genitives. We find that these biases affect gesture order in contexts where no conventionalized system is in place. When a system is in place, participants learn that system, and category‐specific biases do not impact their learning. Our results suggest that different contexts reveal distinct types of cognitive biases; some are active during learning and others are active during language creation.
- Open Access
The lexicon adapts to competing communicative pressures: Explaining patterns of word similarity
- Open Access
Iterated language learning is shaped by a drive for optimizing lossy compression
It has recently been theorized that languages evolve under pressure to attain near-optimal lossy compression of meanings into words. While this theory has been supported by broad crosslinguistic empirical evidence, it remains largely unknown what cognitive mechanisms may drive the cultural evolution of language toward near-optimal semantic systems. Here, we address this open question by studying language evolution in the lab via iterated learning. Across two qualitatively different domains (colors and Shepard circles), we find that semantic systems evolve toward the theoretical limit of efficient lossy compression, and over time, converge to highly efficient systems. This provides direct evidence that adult learners may operate under a bias to maintain efficiently compressed semantic representations. Moreover, it demonstrates how this bias can be amplified by cultural transmission, leading to the evolution of information-theoretically optimal semantic systems.
Open AccessIterated language learning is shaped by a drive for optimizing lossy compression
It has recently been theorized that languages evolve under pressure to attain near-optimal lossy compression of meanings into words. While this theory has been supported by broad crosslinguistic empirical evidence, it remains largely unknown what cognitive mechanisms may drive the cultural evolution of language toward near-optimal semantic systems. Here, we address this open question by studying language evolution in the lab via iterated learning. Across two qualitatively different domains (colors and Shepard circles), we find that semantic systems evolve toward the theoretical limit of efficient lossy compression, and over time, converge to highly efficient systems. This provides direct evidence that adult learners may operate under a bias to maintain efficiently compressed semantic representations. Moreover, it demonstrates how this bias can be amplified by cultural transmission, leading to the evolution of information-theoretically optimal semantic systems.
- Open Access
- 4 January 2025
- Published by California Digital Library (CDL) in Glossa Psycholinguistics
- Vol. 4 (1)
Experimental evidence for semantic and morphophonological productivity in Kîîtharaka noun classes
In nominal classification systems, both form (i.e., morphology, phonology or both) and meaning often interact to determine the class or gender of the noun. In Bantu languages in particular, linguistic analysis has often put the emphasis on meaning, both inherent and evaluative (e.g., diminutive). However, recent quantitative studies have argued that both meaning and morphophonology – the ubiquitous nominal prefixes – serve as cues to class in Bantu, with their robustness and specific aspects potentially differing across individual languages. Here, we conducted an experimental study aimed at establishing whether speakers of Kîîtharaka (Bantu, E54) are sensitive to both semantics and morphophonology when classifying novel Kîîtharaka nouns. We used two wug-task-style experiments to establish whether particular aspects of meaning or form (here, nominal prefixes) would influence participants’ production of agreement on nominal dependents. Results showed that speakers are sensitive to two inherent features, Human and Fruit, and evaluative features like Augmentative, Pejorative and Diminutive. On the other hand, they are robustly sensitive to all morphophonological features tested when classifying novel nouns. Our results suggest that semantic features are generally less productive than morphophonology in the Kîîtharaka nominal classification system.
- 4 January 2025
- Published by California Digital Library (CDL) in Glossa Psycholinguistics
- Vol. 4 (1)
Open AccessExperimental evidence for semantic and morphophonological productivity in Kîîtharaka noun classes
In nominal classification systems, both form (i.e., morphology, phonology or both) and meaning often interact to determine the class or gender of the noun. In Bantu languages in particular, linguistic analysis has often put the emphasis on meaning, both inherent and evaluative (e.g., diminutive). However, recent quantitative studies have argued that both meaning and morphophonology – the ubiquitous nominal prefixes – serve as cues to class in Bantu, with their robustness and specific aspects potentially differing across individual languages. Here, we conducted an experimental study aimed at establishing whether speakers of Kîîtharaka (Bantu, E54) are sensitive to both semantics and morphophonology when classifying novel Kîîtharaka nouns. We used two wug-task-style experiments to establish whether particular aspects of meaning or form (here, nominal prefixes) would influence participants’ production of agreement on nominal dependents. Results showed that speakers are sensitive to two inherent features, Human and Fruit, and evaluative features like Augmentative, Pejorative and Diminutive. On the other hand, they are robustly sensitive to all morphophonological features tested when classifying novel nouns. Our results suggest that semantic features are generally less productive than morphophonology in the Kîîtharaka nominal classification system.
- Open Access
- 1 January 2025
- Published by Open Library of the Humanities in Glossa: A Journal of General Linguistics
- Vol. 10 (1)
Semantic and morphophonological productivity in the Kîîtharaka gender system: A quantitative study
Nominal classification systems are characterized by diverse and sometimes complex gender assignment rules that interact in dynamic ways. The specifics of how gender assignment works can vary widely from language to language. Nominal classification systems of Bantu languages in particular have been the subject of long-standing debate. Traditional accounts of gender assignment are based principally on abstract semantic features which have been found to be unreliable in at least some present-day Bantu languages. Such analyses, in addition to relying on arbitrary sets of semantic features, are problematic, since they provide no clear definition of what it means for a particular feature to be productive. Here we follow a more recent approach, using corpus data to evaluate productivity. We compiled a corpus of 2,327 Kîîtharaka nouns, each coded for a set of semantic and morphophonological features. We then used the Tolerance Principle (Yang, 2016) to measure the predicted productivity of each feature. Our results indicate that morphophonological features are highly productive in Kîîtharaka. However, of the tested semantic features, relatively few were predicted to be productive, including Human, and some evaluative features (Augmentative, Pejorative, Diminutive). The feature Human is in fact only productive for a specific subset of nouns, indicating a possible interaction between features. We discuss why this might be, and highlight the implications of this approach for Kîîtharaka, and for the study of gender assignment in Bantu more broadly.
- 1 January 2025
- Published by Open Library of the Humanities in Glossa: A Journal of General Linguistics
- Vol. 10 (1)
Open AccessSemantic and morphophonological productivity in the Kîîtharaka gender system: A quantitative study
Nominal classification systems are characterized by diverse and sometimes complex gender assignment rules that interact in dynamic ways. The specifics of how gender assignment works can vary widely from language to language. Nominal classification systems of Bantu languages in particular have been the subject of long-standing debate. Traditional accounts of gender assignment are based principally on abstract semantic features which have been found to be unreliable in at least some present-day Bantu languages. Such analyses, in addition to relying on arbitrary sets of semantic features, are problematic, since they provide no clear definition of what it means for a particular feature to be productive. Here we follow a more recent approach, using corpus data to evaluate productivity. We compiled a corpus of 2,327 Kîîtharaka nouns, each coded for a set of semantic and morphophonological features. We then used the Tolerance Principle (Yang, 2016) to measure the predicted productivity of each feature. Our results indicate that morphophonological features are highly productive in Kîîtharaka. However, of the tested semantic features, relatively few were predicted to be productive, including Human, and some evaluative features (Augmentative, Pejorative, Diminutive). The feature Human is in fact only productive for a specific subset of nouns, indicating a possible interaction between features. We discuss why this might be, and highlight the implications of this approach for Kîîtharaka, and for the study of gender assignment in Bantu more broadly.