KARA MORGAN-SHORT (University of Illinois Chicago)

Artificial linguistic systems as tools for understanding adult second and additional language learning

Artificial linguistic systems can offer researchers test tube-like models of adult second and additional language (L2/A) learning through which specific questions can be examined under tightly controlled conditions. This talk reviews artificial linguistic systems that are relevant to L2/A, and it considers what this research has revealed about the mechanisms that underpin L2/A grammar learning. First, the talk considers artificial linguistic systems that consist of structured form(s) but that are devoid of meaning, for example artificial grammars (Buffington et al., in progress; de Vries et al., 2010). Next, the talk will move to artificial linguistic systems that consist of both structured form and meaning, for example, artificial languages and mini-natural languages (e.g., Morgan-Short et al., 2012; Rebuschat et al., 2021). For both types of artificial linguistic systems, the talk will first describe the types of systems that have been utilized in research. Second, the talk will focus on a key study in each area and will present a brief synthesis of findings relevant to L2/A. Finally, the ecological validity of each of these types of systems for adult L2/A will be considered with a call for further validation. Overall, the case will be made that using artificial linguistic systems can serve as an effective and productive way of investigating the development of adult L2/A grammar.

KENNY SMITH (University of Edinburgh)

The relationship between frequency and irregularity in language change: an experimental approach using iterated artificial language learning

The expressive power of natural languages depends on their regular compositional structure, which allows us to express and understand an infinite set of messages. However, we also need to account for irregular exceptions to regular rules, common in natural languages. Historical linguistics has established a correlation between irregularity and frequency, but there is uncertainty as to the mechanisms responsible for this correlation: it has been attributed to preferential irregularisation of frequent items, or preferential regularisation of infrequent items. Being observational, analyses of historical data cannot speak directly to causal mechanisms; however, historical linguistics provides a rich source of hypotheses about mechanisms shaping linguistic systems, which can subsequently be tested in controlled experiments that can speak to causality. I will present an iterated learning experiment where participants learn and reproduce a miniature language across multiple generations, providing an experimental simulation of language transmission. Our results show that the frequency-irregularity correlation can be explained by the relationship between frequency, regularity and learnability, without needing to appeal to frequency-dependent irregularisation. We find that systems of plural marking regularise across generations of transmission, but that high-frequency items remain irregular. Our results further show that the persistence of irregularity is due to high frequency overriding pressures which normally reduce learnability, such as low generalisability of the inflectional strategy (suppletion is disfavoured except for high frequency items) and low type frequency (belonging to a small inflectional class is disfavoured except for high frequency items). This test-case hopefully illustrates the potential to use artificial language learning and iterated learning as tools to test hypotheses from the historical linguistics literature about the individual-level psychological processes driving language change. 

JENNIFER CULBERTSON (University of Edinburgh)

ALL designs for answering complex questions in linguistics

At their inception, artificial grammar learning paradigms were designed to be maximally simple. While that remains an aim that every experimentalist strives for, adding targeted complexity has allowed researchers to push the boundaries of what is possible using ALL. In this talk, I highlight three ways in which more complex ALL designs can shed light on fundamental questions in linguistics. First, by targeting a much more diverse set of populations, we can build much more robust empirical evidence linking common features of language to universal features of the human mind. Second, by integrating transmission (both vertical and horizontal) we can test hypotheses about the specific mechanisms that underlie these links. Third, by manipulating modality we can explore domain-general mechanisms of language emergence. For each of these, I show how the relevant dimension of complexity can be implemented, and summarise some key findings. For the first, I discuss ALL experiments testing whether universal cognitive/perceptual mechanisms drive two typological tendencies around ordering. The impact of diversifying participant populations in these two cases involved very different practical considerations, and led to very different outcomes! For the second, I highlight a case in which including a communicative task in an ALL design called into question a claim about the mechanisms driving efficiency in language. Finally, for the third, I show how using the gestural modality in ALL can help answer fundamental questions about the emergence of linguistic regularity, by combining improvisation and transmission. I end by discussing what I see as one of the next big steps in ALL: designing more linguistically-informed tests to probe exactly what participants learn and how they represent it.

MORA MALDONADO (Université de Nantes)

An artificial learning approach to cross-linguistic regularities in meaning

I will begin by examining the pressures that shape how languages organize categorization systems; specifically, in the domain of personal pronouns. This domain offers a particularly compelling case study, as it involves categorization not in content words, but in functional vocabulary. Next, I will discuss a study on the phenomenon known as the nall lexical gap: the observation that across the world’s languages, the quantificational concept not all (nall) is almost never lexicalized. Through a series of artificial language learning experiments, we explore whether this gap may reflect a cognitive bias against such meanings. Finally, I turn to negative dependencies, beginning with Negative Polarity Items (NPIs)—words that frequently co-occur with negation (e.g., English at all). Because of their restricted distribution, learners may assign different interpretations to NPIs. For example, at all might be interpreted either as a universal quantifier scoping over negation or as an existential quantifier scoping under it. This study investigates how learners interpret NPIs in positive environments, and whether this helps explain why such expressions exist in natural language.

ZARA HARMON (Max Planck Institute for Psycholinguistics, Nijmegen) 

What Kind of Knowledge Supports Semantic Extension? Evidence from Artificial Language Learning

In a series of artificial language learning (ALL) studies, we investigated the cognitive mechanisms underlying creativity in language acquisition and language change. In our initial work, we found that the extension of a form to novel contexts—a process central to linguistic creativity—was influenced by the form’s frequency. Speakers tended to extend frequent forms rather than their less frequent competitors to express new but related meanings. Our design enabled us to test alternative explanations for the role of frequency in semantic extension. However, because token and type frequency were not independently controlled, it remained unclear whether learners were primarily relying on repeated exposure to a single form–meaning pairing (token frequency) or on the diversity of contexts in which a frequent form appeared (type frequency). While token frequency enhances the accessibility of a form given identical or overlapping semantic features, type frequency may help differentiate semantic features consistently predictive of the form’s use from the broader, varying contexts in which the form occurs. To test this, we conducted a follow-up ALL study in which type and token frequency were manipulated independently. Our findings suggest that although token frequency facilitates access to a form in familiar contexts, it is the diversity of contexts—captured by type frequency—that more strongly predicts its extension to novel uses.

GARETH ROBERTS (University of Pennsylvania) 

However, they raise their own difficulties. A particularly common worry about artificial-language paradigms concerns precisely their artificiality. How can interaction with a novel language in the lab tell us much of use about the real and complex linguistic behavior that goes on outside it? Here I argue, with reference to specific work in the paradigm, that this concern is ill-founded. In fact, it is arguably orthogonal to artificial language learning in particular as, on the one hand, there is a great deal of artificiality in experimental research on natural language and, on the other hand, many artificial language experiments involve relatively naturalistic contexts. Indeed, in some contexts they allow a kind of naturalism that is otherwise hard to achieve. Such naturalism is primarily and best achieved by the use of games, especially those in which participants are put into a context in which the artificial language is merely a means to a non-linguistic end. In this talk I will give examples of such experiments and argue for the advantages of this approach to the study of language change in particular.

MARIEKE SCHOUWSTRA (University of Amsterdam) 

From gesture to structure: ALL and language emergence in the visual manual modality

Where does linguistic structure come from? Evolutionary linguistics investigates how individual-level cognitive mechanisms interact with social and cultural processes to shape linguistic conventions. Artificial language learning (ALL) has been recognised as an important experimental tool in this field, but its use extends beyond written and spoken language. In this talk, I present work combining silent gesture — a laboratory paradigm in which participants convey information using their hands and bodies, but no speech —with ALL, resulting in a framework we call Artificial Sign Language Learning. This approach enables us to examine how pressures from individual learning interact with emerging communicative systems. I will discuss key methodological issues and reflect on what we can and cannot learn from this paradigm.