Author: Henry Conklin

How to grow a grammar: Syntactic development in infancy

Laurel Perkins, University of California Los Angeles

Tuesday, June 15 2021, 16:00-17:00 BST
Zoom Details: [Please Request]

What we can learn depends on what we already know; a child who can’t count cannot learn arithmetic, and a child who can’t segment words cannot identify properties of verbs in her language. Language acquisition, like learning in general, is incremental. How do children converge on a grammar for their language using incomplete and noisy representations of their linguistic input?

In this talk, I’ll be looking at these questions through the lens of infants’ wh-dependency acquisition. I’ll present a collection of studies that use behavioral methods to investigate infants’ representations of wh-questions, and computational methods to investigate how those representations are acquired. These studies show: (1) that infants in their second year of life represent moved arguments in wh-questions; (2) that their representations of these questions develop, with local argument relations represented before argument movement; and (3) that learners might identify movement dependencies in their language by combining prior syntactic knowledge with smart statistical learning mechanisms. This case study shows how a multifaceted approach, drawing from formal linguistics, developmental psycholinguistics, and computational cognitive modelling, can inform our theories of grammar acquisition in development.

Showing and seeing: enactment in British Sign Language conversations

Gabrielle Hodge & Sannah Gulamani, Deafness Cognition & Language Research Centre, University College London

Tuesday, June 8 2021, 11:00-12:00 BST
Zoom Details: [Please Request]

It is widely accepted that investigations of enactment (non-conventional, improvised, bodily depictions of events) are integral for understanding the origins and evolution of language (see e.g., Żywiczyński, Wacewicz & Sibierska, 2018). However, there is significant disconnect in how enactment in spoken and signed languages is understood and analysed, which inhibits cross-modal comparability and investigation of the role of deaf signed languages in evolutionary theory. Here we take the position that both signers and speakers use improvised bodily enactment with and without more conventionalised semiotic strategies to mimetically depict the actions, utterances, thoughts and feelings of themselves, other people, animals, and things (Tannen, 1989; Metzger, 1995). Proficient use of enactment in deaf signed language ecologies is vital for understanding others and making oneself understood (see e.g., Cormier, Smith & Zwets, 2013; Ferrara & Johnston, 2014). Indeed, enactment is just one of several strategies for depicting in face-to-face communication, which are tightly integrated with strategies for describing and indicating (Clark, 1996; see also Ferrara & Hodge, 2018). However, unlike with spoken languages (e.g., Hakulinen & Selting, 2005), little is known about signed conversations, and the role of non-conventional semiotics during these interactions. One question is how signers use bodily enactment to visibly depict a referent while indexing other ‘invisible’ referents in the signing space around them. This enables signers to ‘show’ one referent with their body while simultaneously ‘seeing’ another (Winston, 1991; Engberg-Pedersen, 1993; Liddell, 2003). Here we describe how deaf signers of British Sign Language (BSL) do this during dyadic conversations, to highlight the coordinated complexity of depiction and indexicality within enactments occurring in everyday interactions between deaf signers of an established deaf community signed language.

Why is scaling up models of language evolution so hard?

Marieke Woensdregt & Mark Blokpoel, Radboud University

Tuesday, June 1 2021, 11:00-12:00 BST
Zoom Details: [Please Request]

Computational model simulations employing versions of Bayesian rational inference have been very fruitful for gaining insight into how the systematic structure we observe in the world’s natural languages could have emerged through cultural evolution. However, these model simulations operate on a toy scale compared to the size of actual human vocabularies, due to the prohibitive computational resource demands that simulations with larger lexicons would pose. Using computational complexity analysis, we show that this is not an implementational artifact, but instead it reflects a deeper theoretical issue: these models are (in their current formulation) computationally intractable. This has important theoretical implications, because it means that there is no way of knowing whether or not the properties and regularities observed for the toy models would scale up. All is not lost however, because awareness of intractability allows us to face the issue of scaling head-on, and can guide the development of our theories.

Other authors will be in attendance but not presenting

Inductive Biases & Compositional Generalization

Henry Conklin, University of Edinburgh

Tuesday, May 25 2021, 11:00-12:00 BST
Zoom Details: [Please Request]

Neural Networks increasingly perform a wide array of Natural Language Processing (NLP) tasks with high-fidelity. Despite this there is little evidence of their ability to generalize compositionally, or robustly outside of their training data. This disconnect between task performance and systematic generalization may best be explained as a result of underspecification. In supervised learning training data alone may not adequately specify for compositional strategies, or strategies that generalize robustly. We present experiments looking at a meta-augmented form of supervised learning as a way to incorporate prior bias into training and mitigate some issues of underspecification. We show how introducing certain kinds of bias motivated by human cognitive constraints can aid generalization performance on two different compositional generalization tasks. More broadly these experiments show how biases introduced during training could be used to condition the kinds of generalization strategies that emerge. In future we hope that introducing more biases analogous to those found in humans results in neural models that can arrive at more human-like solutions.

Cultural Evolutionary Psychology

Cecilia Heyes, University of Oxford

Tuesday, May 18 2021, 11:00-12:00 BST
Zoom Details: [Please Request]

Cultural evolutionary psychology seeks to explain the origins and effectiveness of distinctively human cognitive mechanisms by combining the resources of cognitive science and evolutionary theory. In contrast with classical evolutionary psychology, it suggests that these mechanisms have been shaped primarily by culture; by Darwinian selection operating on socially rather than genetically inherited variants. In other words, cultural evolutionary psychology casts distinctively human psychological mechanisms as ‘cognitive gadgets’ rather than ‘cognitive instincts’, but it is not a blank slate theory. During human evolution, often via Baldwinisation, genetic selection has tuned motivational, attentional, and learning processes that we share with other animals to make our developing minds more malleable by information from other agents. Using morality, imitation and metacognition as examples, I will sample the evidence from developmental psychology, comparative psychology and cognitive neuroscience that supports cultural evolutionary psychology and discuss the opportunities and challenges it presents for those who want to understand not only how our minds work, but why they work that way.

Chaining and the growth of word meaning

Yang Xu, University of Toronto

Tuesday, May 11 2021, 16:00-17:00 BST
Zoom Details: [Please Request]

Natural language relies on a finite lexicon to express a potentially infinite range of meanings. This tension creates a funnel effect where meanings are compressed through a limited set of words. Prior work suggests that word meanings are structured for efficient compression. I describe recent development that extends this work to investigate the cognitive mechanisms in the dynamic growth of word meaning through time. I first present work that synthesizes cognitive linguistic theories of chaining with classic models of categorization to predict the historical extension of numeral classifiers for emerging referents. I then present evidence that similar models of chaining predict children’s spatial word generalization. Our findings suggest that an exemplar-based model of chaining may underlie the general mechanisms in word meaning growth. I discuss applications of this work to natural language processing and implications for research in lexicon evolution.

Differential argument marking and efficiency: New typological and experimental evidence

Natalia Levshina, Max Planck Institute for Psycholinguistics

Tuesday, May 5 2021, 11:00-12:00 BST
Zoom Details: [Please Request]

In this talk I want to discuss differential object marking as a manifestation of efficiency in language (Gibson et al. 2019). I will present the results of two experiments based on artificial languages. One involves only learning, and the other one is a communication game. This evidence supports the idea expressed by Smith and Culbertson (2020) that efficient differential marking emerges in communication, rather than in simple language learning. Moreover, the data demonstrate that differential marking can emerge solely for the purpose of distinguishing between the arguments, excluding the other motivations proposed in the literature. I will also present new cross-linguistic data that support the efficiency-related interpretation of differential marking.

Learning biases in paradigmatic and syntagmatic morphological relations

Carmen Saldana, University of Zurich

Tuesday, Apr 27 2021, 11:00-12:00 BST
Zoom Details: [Please Request]

Despite the astonishing variation that morphological structure displays cross-linguistically, there are certain patterns that reappear across many languages and others that are very rare or even unattested. My research focuses on exploring whether these cross-linguistic asymmetries can be explained by shared features of human cognitive systems. In this talk, I will present a series of artificial language learning studies which aim to test this hypothesised link in morphological structure both at the syntagmatic (i.e., morpheme order) and paradigmatic level (e.g., structure of inflectional paradigms). I will focus in particular on three cross-linguistic regularities: 1) affixes that belong to the same category tend to appear in the same syntagmatic slot, 2) affixes with stronger (structural) relationships to the stem tend to appear closer, and 3) patterns of syncretism in inflectional paradigms tend to reflect more natural partitions of the morphosemantic space. The results from the studies I will present suggest that these regularities across the languages of the world may be shaped by universal features of human cognition during language learning and transmission.

Adjective Ordering Preferences: From Conceptual Roots to Linguistic Strings

Monica Do, The University of Chicago

Tuesday, Apr 20 2021, 16:00-12:00 BST
Zoom Details: [Please Request]

When a noun is modified by more than one adjective, there is typically a preferred order in which adjectives appear (Dixon, 1982). Speakers follow this order intuitively, whether they speak a pre-nominal language like English (e.g., “red leather jacket”) or a post-nominal language like Vietnamese (e.g., “jacket red leather”). The current work asks where do these Adjective Ordering Preferences (AOPs) come from and how they interact with the process of uttering a multi-adjective string? In an initial set of studies in English, I first provide evidence that AOPs in reflect biases in speakers’ non-linguistic conceptual representation of the entity they are talking about. I then show how factors related to lexical accessibility of an adjective can modulate biases in speakers’ conceptual representations of objects and their attributes during real time language production. Together, these studies (i) shed light on the psychological underpinnings of AOPs; (ii) bridge domains of research that have historically been pursued completely independent of each other; and (iii) further our current knowledge of the relationship between thought and language.

Suggested Readings:
1. Papafragou & Grigoroglou (2019): This paper does not pertain to adjectives, specifically, but provides a broad overview of the processes involved in language production processes, as well as the relationship between linguistic and conceptual representations that I assume in my work.

2. Trotzke & Wittenberg (2019): This recent overview article introduces other work that looks at the question of AOPs from a more theoretical perspective. For our purposes, it provides a brief introduction to the problem presented by AOPs and a glimpse into the types of issues under investigation in the AOP literature.

The neural basis of gesture-speech integration and interaction during online comprehension

Yifei He, Translational Neuroimaging Lab, Department of Psychiatry and Psychotherapy, Philipps University Marburg

Tuesday, Mar 30 2021, 11:00-12:00 BST
Zoom Details: [Please Request]

Human daily communication is realized in a multimodal manner. Besides auditory speech, visual input such as hand gestures also plays an important role. Despite advances in neuroscientific investigations on language processing, we know relatively little about the neural basis of how gesture integrates and interacts with speech during online comprehension. In this talk, I will present evidence from EEG, fMRI, and simultaneous EEG-fMRI, showing the brain dynamics of how the two input channels are integrated as coherent semantic representations. I will also present studies on how gesture impacts the semantic processing of language (auditory speech & visual sentences). i) EEG evidence from a controlled experiment shows that the social aspects of gesture (body orientation) may directly influence the N400 amplitude during sentence processing. ii) fMRI studies employing naturalistic paradigms suggest that gesture facilitates the neural processing of natural speech; more specifically, this facilitation may be realized as reduced brain activation on semantic prediction but not prediction error on a single word level.