UNIVERSAL · A Universal Model of Word Comprehension
FP7 — People (Marie Curie Actions)
- Duration
- 2014-05-01 → 2016-04-30
- EU contribution
- €166,336
- Participants
- 1
- Scheme
- MC-IIF
Lines connect the coordinator with its partners.
Results in brief
A Universal Model of Word Comprehension
The aim of this project was to evaluate the cross-linguistic differences associated with visual and auditory word recognition, and how these representations map onto word meaning. Overall, this work has revealed that a number off prior published models and theories are not theories of language in general, but rather are theories of English in particular. This work has helped to highlight important aspects of the visual and auditory word recognition systems that are flexible and are able to adapt to the specific constraints imposed by each language. For example, we have shown how cross-linguistic differences in the number of anagrams in Hebrew versus in English explain why differential sensitivity emerges for identifying the position of letters in each language. We have also shown that the degree of information flow between auditory and visual word representations is mediated by the spelling-sound and sound spelling consistency between the languages, as assessed in English, French, and Spanish. In addition to this work on cross-linguistic differences, we have also conducted two main extensions of the original proposal. One of these extensions was to connect our work on language with the broader research on statistical learning of regularities across space and time. This work has helped highlight now the differences that emerge between visual and auditory word processing may not be due to language processing per se, but may instead be attributable to domain-general processing mechanisms that are interacting with modality specific constraints. Furthermore, this project has helped connect studies of language to computational models not only at the behavioral level, but also at a more detailed and mechanistically explicit level in terms of the online measurement of the neural correlates of language processing. Consequently, the resulting theories and models offer an unprecedented unifying account of various aspects of language processing, which are suitable for integration of future findings from a wide area of research and for extensions to other related systems (e.g., episodic memory, semantic memory).
Data: CORDIS, © European Union
Project objective
The proposed research will develop a “universal” connectionist architecture and a set of models capable of handling key cross-linguistic differences across English, Spanish, and Hebrew. This work is complemented with targeted empirical investigations.Computational Investigations. I will develop improved connectionist models which employ more general architectural assumptions and that are able to learn to comprehend words from languages that impose different division of labour between orthography, phonology, and semantics. These models will explain and generate predictions regarding how a general architecture and the statistical properties of a language give rise to a range of cross-linguistic differences reported in the behavioural literature. This will be accomplished by combining a model of English, Spanish, and Hebrew with a response generation model to simulate performance in a range of tasks (e.g., differential transposed-letter priming in English & Spanish vs. Hebrew; stronger syllable priming effects in Spanish vs. English; greater semantic priming in naming Hebrew vs. English and English vs. Spanish). Additionally, the models will be developed within a biologically-plausible connectionist framework to connect the model to the ERP literature (Armstrong & Plaut; 2013; Laszlo & Armstrong, 2013).Behavioural & pilot EEG/ERP Investigations. To evaluate the modeling work, behavioral (and pilot electrophysiological) data will be collected from masked priming lexical decision tasks designed to engage orthographic, syllabic, and semantic representations. Language development will also be studied in pilot behavioural experiments with children.I will develop and release a public implementation of my empirically-validated models and data. These models will account for a broad range of existing effects, generate predictions for new experiments, and be readily extendable to related areas (e.g., modeling bilingualism and language impairments).
Original text from CORDIS.
Participants
- BCBL BASQUE CENTER ON COGNITION BRAIN AND LANGUAGE · San SebastianCoordinatorSpain
Links
Data: CORDIS, © European Union