skip to content

Cambridge Language Sciences

Interdisciplinary Research Centre

Presenting work by early-careers researchers at the University of Cambridge

Online registration has now closed. If you missed the deadline but would like to request a place please contact Hazel ( or Jane ( before Monday 18th.

Open to all language sciences researchers at the University of Cambridge, and supported by Cambridge Language Sciences IRC.


Registration & coffee

13.55 Welcome (Dr Andrew Caines)
14.00 Dr Anastasia Klimovich-Gray (Dept. of Clinical Neurosciences)

Dr Paul Nulty (CRASSH)


Poster exhibition & refreshment break

16.00 Dr Dieuwerke (Dee) Rutgers (Faculty of Education & MEITS project)
16.30 Alexander Kuhnle (Dept. of Computer Science and Technology)
17.00 Dr Ana Pérez (Linguistics, Modern & Medieval Languages)

Closing remarks

17.45-19.30 Drinks & canapés



Domain-general vs. domain-specific in spoken word and phrase recognition (Anastasia Klimovich-Gray)

Language comprehension invokes cortical computations that have been broadly separated into two camps: domain-general, i.e. those that support processing that is not specifically linguistic in nature; and domain-specific, i.e. those that are uniquely linked to aspects of linguistic processing. Critically it is yet undetermined whether such strict computational dichotomy is valid and if so, what cortical networks support what computations.

Here we present cross-linguistic evidence from two fMRI studies, one in Russian and another in English. Cortical activity was recorded as participants listened to spoken utterances varying in the degree of their domain-general or domain-specific complexity. To assess which areas of the bilateral frontotemporal language processing network predominantly support domain-general or domain-specific computations, we used a variety of statistical techniques - univariate subtractions, connectivity measurements, regression and multivariate analyses. Results from the Russian study show that the left frontotemporal network supports demands of combined domain-general and domain-specific complexity, routinely present in morphologically complex Russian words. Results from the English study show increased bilateral frontotemporal connectivity for domain-general aspects of complex word processing. The combined results imply functional versatility of language-processing networks in their adaptation for the demands posed by different languages.

Semantic Network Analysis of Contested Political Concepts (Dr Paul Nulty)

Political concepts are often characterised as "essentially contested" in the sense that their essential meanings are necessarily in dispute when we deploy them in adversarial political discourse. When tasked with pinning down an elusive word meaning, modern lexicographers, computational linguists, and natural language engineers usually turn to a descriptive analysis of the term's use in context, often looking for statistical patterns of syntactic or document-based word co-occurrence in large collections of digital text. This talk presents an application of the tools of statistical corpus semantics to the problem of delineating the various complex, multi-faceted, and value-laden meanings of abstract political concepts. As an example, comparisons of the lexical-semantic structure of political terms from libertarian and socialist online communities are presented. Network analysis methods including community-detection and measures of centrality are used to compare and contrast the terms of particular importance to the structure of word association networks from different ideologies. 


Putting CLIL into practice in multilingual primary education settings: Towards joint conceptualisations and successful implementations (Dr Dieuwerke Rutgers)

In response to continued internationalisation, the number of primary schools in Europe offering early foreign language (FL) programmes, particularly with English, has risen sharply in recent years. Increasingly, schools are adopting a content-and-language-integrated or CLIL approach to FL teaching, whereby the teaching of subject content (e.g. science) and language (e.g. English) occur simultaneously. This places new demands on primary school teachers associated with the integration of content and language didactics.

One the one hand, this relates to teachers’ understanding and implementation of general CLIL principles and procedures, which continues to be subject of debate and research, particularly in terms of the conceptualisation of integration within CLIL (Nikula et al., 2016). On the other hand, this relates to the implementation of CLIL procedures specific to the (developmental) age group that primary teachers teach, particularly as CLIL has so far been more widely established and researched at the secondary level. Moreover, globalisation has also contributed to an increase in the cultural and linguistic diversity of classrooms. This poses further challenges for teachers associated with the incorporation of more multilingual and culturally-inclusive approaches to CLIL, and the successful integration of minority children.

Thus, primary schools and teachers aiming to successfully implement CLIL lessons not only face the challenge of translating secondary CLIL research and practice to the primary context, but also of having to respond adequately and effectively to the rapidly changing linguistic landscape in their classrooms, in relation to both of which established CLIL theory and practice only provides partial answers. In this talk, I will discuss ongoing research on teacher professionalisation for primary CLIL in the Netherlands, presenting findings from an initial research phase dedicated to identifying the principles of and current gaps in CLIL teaching in multilingual primary settings, as well as implications and current directions.  

“Unit-testing" deep learning with synthetic data for more informative evaluation (Alexander Kuhnle)

Deep neural networks, in combination with huge datasets, have transformed the field of Natural Language Processing. It is common practice nowadays to rely on standard benchmarks to investigate and compare system performance. However, I will argue by example of the visual question-answering task that this led to a wide range of systems with little indication of their strengths and weaknesses. One reason is that datasets are often not targeted at specific phenomena but at rather general abilities, like multimodal reasoning. Moreover, many datasets have been shown to contain various biases and correlations which systems use to obtain good performance without actually learning anything interesting. Next, I will present a complementary evaluation approach based on automatically generated abstract data, which makes it possible to tailor the evaluation to a model's specific improvements and investigate its behaviour in detail. Compared to real-world datasets, which act as application benchmarks, such synthetic data is more akin to unit-testing: defined on an abstract level and targeted at a specific capability. Finally, I will discuss some common reservations against synthetic data in NLP and emphasise its potential relevance particularly in the context of deep learning.

Multisensory semantic integration in inferential comprehension (Dr Ana Pérez)

Although language comprehension is usually multimodal (e.g., listening to speech while observing body gestures), only few studies have investigated the role of multisensory semantic integration in language processing (Willems, Özyürek, & Hagoort, 2009), where, in general, multisensory integration of semantically consistent stimuli is facilitated, whereas it is disrupted with semantically inconsistent stimuli. In the present study, twenty-eight young adults (M age= 21.54) listened to short stories prompting an inference (e.g., “bear”). Subsequently, they were visually presented with either a consistent picture (“bear”) or inconsistent but plausible picture (“penguin”). A final, auditorily-presented sentence was either the expected or the unexpected target word (“bear” vs. “penguin”). Preliminary results show electrophysiological differences in both the picture and target word, indicating difficulty to integrate auditory information that failed to match visual information that agreed with the original interpretation. Our behavioural results (accuracy in a final question) confirmed participants’ comprehension has been disrupted after the presentation of multisensory stimuli that were semantically inconsistent. 


Language choice for emotional inner speech among international students in the UK: an acculturation perspective (Wing Wu, Faculty of Education)

Implicit learning and production of novel prepositions (Giulia Boloventa, Theoretical & Applied Linguistics)How clever are the models exhibiting "super-human" performance on the VQA datasets? (Alexander Kuhnle, Ann Copestake, Dept. of Computer Science & Technology)Italian L1-attrition (Alexander Allan Cairncross, Theoretical & Applied Linguistics)Adversarial post-specialisation of word embeddings (Edoardo Maria Ponti, Theoretical & Applied Linguistics)Constructing an ideal L3 self: a classroom intervention for L3 learners in China (Tianyi Wang, Faculty of Education)Subject to constraints: computational semantics and the syntax of arguments in Ancient Greek (Rachel Grewcock, Faculty of Classics)“He sounded like a person of colour” – An explorative case study of secondary Austrian EFL learners’ perceptions of postcolonial Englishes (Julia Jakob, Faculty of Education)Understanding language teacher cognition: critical reflection as an agent of change (Edsoulla Chung, Faculty of Education)The complex construction of multilingual identity: exploring the dynamics between self and context in the process of third language acquisition in four European sites. (Harper Staples, Faculty of Education)

Predictive neural mechanisms in spoken word recognition and learning (Yingcan Wang, Becky Gilbert, Rik Henson, Matt Davies; MRC Cognition & Brain Sciences Unit)

Flexible meaning: the neuromodulation of noun meaning by a prior verb (Bingjiang Lyu, Alex Clarke, Hun Choi, Lorraine K. Tyler; Centre for Speech, Language & the Brain)

Tuesday, 19 June, 2018 - 13:30 to 19:30
Contact name: 
Jane Walsh
Contact email: 
Event location: 
Cavonius Centre, Gonville & Caius College

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.