Final Activity Report Summary - BIT (The Bayesian Information-Theoretical Model of Language Processing in the Brain)
The main objective of this project was the extension of the information theoretical model of lexical comprehension to account for orthographic processing. This permits to have, for the first time, an integrated mathematical account of lexical recognition. The project had two main components, one computational/mathematical and the other experimental in nature:
The computational goal was to extend a previously proposed model of morpho-semantic processing to account also for the processing visual and orthographic information (which was not modelled in the previous version). In this respect, the model has been completed, a first version of the model was published (Moscoso del Prado, 2008; Proc.30th Ann. Conf. Cog. Sci. Soc.) and a more complete journal version is submitted. A crucial property of this model is its analytical nature, instead of simulating the process of lexical recognition, and repeating the process to finally get an estimate of the model's predictions, the predictions are directly calculated mathematically, and its results are in agreement with experimental data. This analytical approach constitutes a major departure from the traditional simulation-driven tradition in the cognitive sciences. A further, unplanned, consequence of the modelling effort has been the development of a model of reading for patients suffering from ocular macular degenerations (a visual impairment very common in aged people). This arose as a collaboration with a neighbouring team of visual neuroscientists, in applying the same information theoretical approach to the patient data they had collected. A preliminary version of this model has been published (Bernard, Moscoso del Prado, Montagnini, and Castet, 2008; Proc. 2nd French Conf. Comp. Neurosci. Sci. Soc.) and a further elaborated version of the model is currently being written. As a spin-off project, this has led to a project investigating how the model's predictions of optimal reading strategies given different shapes of the macular degenerations can be applied to develop remedial strategies in improving these patients reading abilities. Finally, a further line of modeling research that has started as a consequence of the a-thematical work in this project is a full-scale investigation of the dynamical properties of human and animal responses, which constitutes the main area of work of the candidate. Finally, a refinement of the morpho-semantic aspects of the model, enabling its integrations with corpus derived information (i.e. the information that can be automatically extracted from texts, in a similar way to the way people extract the information from text and speech) was also developed and produced a journal publication.
The experimental work included EEG and fMRI experimentation to investigate how the information-theoretical measures developed above relate to the patterns of neural activation during lexical processing. Three EEG experiments were performed, and they show that, as was predicted, the measured informational content of words and morphemes is linearly reflected in the power of the EEG signal while processing words. These results are currently being written down. A further line of behavioural experimental work was carried out with research collaborators in Serbia and the US, to investigate the interplay of informational measures in different languages (English and Serbian) and research in further languages (French and Arabic is in progress). This part has led to two journal publications and one more in preparation. An additional area of experimental work has been to extend these ideas from language comprehension into language production, which has led to one journal publication and a further one in preparation.
The computational goal was to extend a previously proposed model of morpho-semantic processing to account also for the processing visual and orthographic information (which was not modelled in the previous version). In this respect, the model has been completed, a first version of the model was published (Moscoso del Prado, 2008; Proc.30th Ann. Conf. Cog. Sci. Soc.) and a more complete journal version is submitted. A crucial property of this model is its analytical nature, instead of simulating the process of lexical recognition, and repeating the process to finally get an estimate of the model's predictions, the predictions are directly calculated mathematically, and its results are in agreement with experimental data. This analytical approach constitutes a major departure from the traditional simulation-driven tradition in the cognitive sciences. A further, unplanned, consequence of the modelling effort has been the development of a model of reading for patients suffering from ocular macular degenerations (a visual impairment very common in aged people). This arose as a collaboration with a neighbouring team of visual neuroscientists, in applying the same information theoretical approach to the patient data they had collected. A preliminary version of this model has been published (Bernard, Moscoso del Prado, Montagnini, and Castet, 2008; Proc. 2nd French Conf. Comp. Neurosci. Sci. Soc.) and a further elaborated version of the model is currently being written. As a spin-off project, this has led to a project investigating how the model's predictions of optimal reading strategies given different shapes of the macular degenerations can be applied to develop remedial strategies in improving these patients reading abilities. Finally, a further line of modeling research that has started as a consequence of the a-thematical work in this project is a full-scale investigation of the dynamical properties of human and animal responses, which constitutes the main area of work of the candidate. Finally, a refinement of the morpho-semantic aspects of the model, enabling its integrations with corpus derived information (i.e. the information that can be automatically extracted from texts, in a similar way to the way people extract the information from text and speech) was also developed and produced a journal publication.
The experimental work included EEG and fMRI experimentation to investigate how the information-theoretical measures developed above relate to the patterns of neural activation during lexical processing. Three EEG experiments were performed, and they show that, as was predicted, the measured informational content of words and morphemes is linearly reflected in the power of the EEG signal while processing words. These results are currently being written down. A further line of behavioural experimental work was carried out with research collaborators in Serbia and the US, to investigate the interplay of informational measures in different languages (English and Serbian) and research in further languages (French and Arabic is in progress). This part has led to two journal publications and one more in preparation. An additional area of experimental work has been to extend these ideas from language comprehension into language production, which has led to one journal publication and a further one in preparation.