skip to content

Cambridge Language Sciences

Interdisciplinary Research Centre

Research based on Experiments with Artificial Languages Symposium

In collaboration with Chuo University in Japan, Cambridge Language Sciences would like to invite you to an afternoon of talks and informal networking for researchers interested in using artificial languages as part of their research.

This hybrid seminar will showcase findings of a Cambridge-Chuo research collaboration on the use of artificial languages to better understand second language learning. It is a free event and is open to language scientists of all disciplines from the University of Cambridge and Chuo University.


John Williams, Professor of Applied Psycholinguistics, University of Cambridge

Shigenori Wakabayashi, Professor of Applied Linguistics, Chuo University 

Junya Fukuta, Associate Professor in Psycholinguistics at Chuo University, Principal Investigator 

Takayuki Kimura, Assistant Professor of Linguistics, Utsunomiya University

Patrick Rebuschat, Professor of Linguistics and Cognitive Science, Lancaster University

András BárányResearch Associate, Bielefeld University (Germany) 

Michelle Sheehan, Professor of Linguistics, Newcastle University

Jennifer Culbertson, Professor of Linguistics and English Language, Edinburgh University

Yuyan Xue, Ph.D. Student, University of Cambridge

Boping Yuan, Professor Emeritus, University of Cambridge


To know more about the collaboration with Chuo University and the research project please visit this page.


Please book your place here. Please note this event is now full to attend in person. Please complete the form if you would like to join online and we will send you the zoom link.




Testing linguistic theories using artificial languages

Jennifer Culbertson


There is now a growing body of research using artificial language experiments to test hypotheses from theoretical linguistics. This is a bold move in a number of respects. Artificial languages are unlike natural languages in many ways, both in terms of complexity and in terms of context. Similarly, artificial language *learners* are unlike natural language learners, and are often adults who bring a wealth of cognitive resources and experience to the task. On top of this, theoretical linguistics research often assumes very specific machinery and representations that are non-trivial to identify psychologically. Nevertheless, I will argue that artificial language experiments have opened the door not just to new sources of evidence for (and against) particular theories, but also to important links with broader research in cognitive science. To illustrate this, I will briefly review three recent projects which put linguistic theories to the test. The first project tests whether artificial language learners are sensitive to a proposed universal syntactic hierarchy in the nominal domain. The second project tests the psychological reality of semantic primitives in the domain of person reference. The third project compares the predictions of different theories for morpheme ordering preferences. Importantly, these projects highlight the specific challenges of using ALL, but also provide some general strategies for dealing with them. In each case, an initial set of results opens the door for further enquiry---both theoretical and experimental.



Incidental learning of attested and unattested ditransitive alignment types

András Bárány, Michelle Sheehan, John Williams


Languages vary in the expression of theme and recipient arguments in ditransitive constructions (cf. English “I gave the flowers to Mary” and “I gave Mary the flowers”). Languages in which the verb can agree with the subject and one object, also show variation with respect to whether the theme or the recipient controls agreement on the verb. But not all logically possible combinations of case-marking (flowers to Mary vs. Mary the flowers) and agreement (with the recipient vs. with the theme) are attested. We present the results of an artificial language learning experiment in which German and English first language participants were exposed, in an incidental learning task, to artificial languages that either matched or did not match attested ditransitive patterns in natural languages. Our hypothesis was that the attested agreement pattern would be more learnable, reflecting sensitivity to universal constraints.


Universality and cross-linguistic influences on the acquisition of unconscious linguistic knowledge

Junya Fukuta, Takayuki Kimura, John Matthews, John Williams, Yuyan Xue, Boping Yuan, Shigenori Wakabayashi


This study investigates sensitivity in adult second language acquisition to constraints on a syntactic operation using a semi-artificial language, exerting control over language experience (e.g., input distribution and frequency) and manipulating learners first language. We tested sensitivity to a universal constraint on extraction from classifier phrases (numeral classifier noun) of the type found in Japanese  extraction of a noun phrase out of a classifier phrase (numeral quantifier floating, NQF) is permitted if the phrase is in a subject/object (i.e., argument) position, but not if it is present within a locative phrase (i.e., adjunct). We created artificial languages combining either English, Chinese, or Japanese lexis with novel morphemes for classifiers and case markers. During training, participants with each of those L1s were exposed to sentences using their native lexis and that contained scrambling, but no extraction of noun phrases from classifier phrases. In a subsequent surprise grammaticality judgement task participants were presented with sentences with completely new word orders and were asked how likely they were to be grammatical in the language that they had just been exposed to. Our focus is on whether sentences that violate the NQF constraint are less likely to be endorsed than other examples of NQF or scrambling, and whether sensitivity to this constraint depends on the participants first language. Hence, we do not test for acquisition of the target rule as such, but rather we create an artificial language environment in which we can test for application of an unconscious grammatical constraint, as may derive either from universal grammar, or the first language, depending on the group involved.


What can cross-situational statistical learning tell us about bilingual development and second language acquisition?

Patrick Rebuschat (Lancaster University and University of Tübingen)


Statistical learning, essentially our ability to make use of statistical information in the environment to acquire (linguistic) knowledge, plays a fundamental role in how we learn languages. Following the seminal work of Saffran et al. (1996), there is substantial empirical evidence demonstrating that infants, children, and adults can rely on statistical learning to complete a variety of linguistic tasks, from speech segmentation and phonological categorization to word learning and syntactic development (see Frost et al., 2019, for a recent re- view). Statistical computations can be applied to a range of language units, including speech sounds, syllables, lexical categories, and syntactic phrases, but they are not limited to the domain of language. Instead, as previous research has shown, statistical learning is domain-general, i.e. it enables us to acquire information from multiple cognitive domains (language, music, etc.) and across a range of modalities (auditory, visual, tactile, etc.) (e.g., Frost et al., 2015). Moreover, statistical learning is not unique to human learners, as non- human primates rely on statistical learning, too (e.g., Rey et al., 2019).

In this talk, I will review recent statistical learning research conducted collaboratively in our group, Lancaster’s Language Learning Lab. The focus will be on experimental studies using the cross-situational learning paradigm developed by Monaghan et al. (2015). In this paradigm, participants are exposed to a novel language in ambiguous contexts under incidental learning conditions. That is, participants face the challenge of having to rapidly map novel sounds or sound sequences to multiple referents in the environment without prior information of the learning target and without feedback. To accomplish this task, participants need to be able to keep track of co-occurrence statistics across multiple learning trials, hence cross-situational statistical learning.

In a sequence of studies, we explored cross-situational learning of novel phonology, words, morphology and grammar, either separately or simultaneously, using either natural languages (Latin, Japanese) or artificial languages (based, for example, on Japanese, Portuguese, German), comparing incidental or intentional learning conditions, the effect of instructional manipulations (e.g., feedback, explicit instruction, spacing) and the role of individual differences (e.g., declarative and procedural memory, working memory). Most of our studies have focused on adult participants (e.g., Monaghan et al. 2019, 2021; Rebuschat et al., 2021; Walker et al., 2020), but we have recently completed studies testing cross-situational learning in children. I will conclude the presentation with a reflection on the implications of this research for the study of bilingual development in children and adolescents and second language acquisition in adults.


For any queries, please email us at Language Sciences Events


Friday, 15 March, 2024 - 14:00 to 18:00
Event location: 
SG02 (ground floor), Alison Richard Building

What we do

Cambridge Language Sciences is an Interdisciplinary Research Centre at the University of Cambridge. Our virtual network connects researchers from five schools across the university as well as other world-leading research institutions. Our aim is to strengthen research collaborations and knowledge transfer across disciplines in order to address large-scale multi-disciplinary research challenges relating to language research.