Effects of early acquisition, experience, and impairment on language processing: A computational model of reading through the lifespan

Lead Research Organisation: Lancaster University
Department Name: Psychology

Abstract

Early experience has a profound impact on cognitive and social development throughout the lifespan. Understanding these effects of early experience first requires uncovering the cognitive consequences of this experience on the way in which the human mind develops. In this project we test the influence of early experience on development of language, and the implications of the trajectory of early learning on language processing across the lifespan. We use computational models to align theoretical models of early experience affecting processing with behavioural observations, and track the effect of life experience on cognitive processing from acquisition through all stages of life.
We focus on the acquisition of words, because there is a large body of empirical literature on word processing at all life stages, well-established theoretical models of how words are processed in the cognitive system, as well as several computational implementations of word processing which facilitate direct comparisons between theories. In word processing, there is evidence that the "age of acquisition" (AoA) of a word has a profound impact on its storage and retrieval, such that early acquired words tend to have more associations with other words, and tend to be retrieved more quickly. Early words tend also to be more resistant to impairment as a consequence of neurodegeneration such as the consequences of Alzheimer's Disease.

This project will implement and test large-scale computational models of word processing that, as an innovation, introduce realistic properties of human language exposure to discover their consequences on how words are represented and processed. The model builds on previous connectionist models of reading, which are particularly appropriate for simulating development because they enable gradual accretion of knowledge over time, as a consequence of environmental exposure.

The models enable us to compare alternative theories about the locus of AoA in language - whether it is in the representations of words' meaning, or whether it is in the learning of mappings among meaning, sound, and written forms of words. The models also permit us to track whether word processing alters with age, by monitoring the involvement of meaning in the model over exposure simulating lifelong exposure to words. The models also enable us to test whether effects are generalisable to other languages, by testing the patterns of AoA in Spanish and English. Further, we can simulate cognitive impairment in the models and relate the effects of early experience in the impaired model to word processing in patients with degenerative diseases.

The modelling provides valuable theoretical insights into the way in which lifelong experience affects cognition. These insights in turn have potentially important impact in terms of understanding the effects of variability in early experience on social and cognitive development, as well as the factors affecting ageing and brain disease in cognitive performance. Thus, the models provide an important step towards proposing and testing the validity and effectiveness of interventions and protective factors against detrimental effects of initially reduced language exposure, and later language degeneration.

Planned Impact

The proposed research has particular impact for education and health, having implications for quality of life, and addressing the ESRC's strategic priority of Influencing Behaviour and Informing Interventions. Our impact pathway ensures that the research in early influences on learning have outcomes in terms of societal impact.

Impact for Education
The project has impact potential for understanding the role of early experience and early language exposure for language processing across the lifespan. This emphasis on early experience has implications for predicting relative ease and difficulty of vocabulary acquisition as a consequence of previous learning. We will ensure impact of this research by:
Communicating with partners in Educational Authorities, teachers and speech therapists, NHS research network directors - via the NorthWest NHS Medicines for Children Research Network, and educational charities. These partners are already in contact with Lancaster University's Centre for Research in Human Development and Learning, as Advisors to this Centre, of which the PI is the Centre Director.
These partners have already been informed of the research, and as the project progresses presentations on the project's research will be given to this panel on an annual basis. Interacting with these partners will align applications in policy and practice with the research results, and this dialogue will assist in informing the research direction in terms of designing effective interventions and models of individual learning.
We have also approached a children's book publisher, but do not yet have formal agreement to involvement in the research project. As the project proceeds, we will engage a publisher partner to ensure that commercial and educational practices are informed by our research on early language acquisition.

Impact for Health
We have identified partners that are health practitioners and policy makers via connections to the Centre for Ageing Research at Lancaster University. These potential beneficiaries of the research include UK charities addressing challenges of ageing, and NHS representatives interested in ageing and neurodegeneration. We have established initial interest in the aims of our project, and we will increase this involvement to result in impact evidence through regular presentations and dialogues with these practitioners, through fora organised by the Centre for Ageing Research as well as Lancaster University's School of Health and Medicine annual events drawing together health policy makers and practitioners with academics. As with the educational partners, establishing the users' needs and priorities and using these to inform our research will lead to maximising impact.

Public engagement
The research is also of general interest and relevance to the broader public, particularly care-givers interested in the effect of early experience on language learning. Our impact plan includes regular press releases on the results of the studies, as well as outreach events in public understanding of science events. We will contribute to the British Science Festival, the Manchester Science Festival, and to local science organisation events, such as Café Scientifique.

Web-based presence
In addition to user-tuned research presentations and talks at public engagement events, we will also provide research reports and lay summaries of our project's activities that are accessible to various user groups - care-givers, educational and health practitioners, and commercial publishers.
 
Description This grant is now completed, we have met all the targets for the intended research programme.
We have developed a computational model of reading, involving representations of written, spoken, and meaning of words. We have trained it to mimic gradual exposure to a growing vocabulary, and reproduced in the model the key psycholinguistic behavioural effects of age of acquisition of words and their effect on reading. The first report on this work is published as a 6 page paper to the Cognitive Science Society Conference, and the second is submitted as a journal publication. We have extended this work to develop alternative representations of meaning in the model, to determine the extent to which the key psycholinguistic observations are due to particular choice of representation. The results of this alternative model of meaning confirm that the observed effects do not depend on our choice of representations.

We have also developed a model of reading in Spanish, for comparison with the model of English, and related this to psycholinguistic data provided by our international collaborator Fernando Cuetos.

We have also achieved three additional spin-off projects.
First, we have measured the effect of lifelong exposure to the effect of frequency on reading. Low frequency words tend to be read faster and faster as a reader ages. Our model shows that this is a consequence of increased exposure to these rare words in the reader's environment. We therefore have a model of lifelong reading explained in terms of cognitive processing interacting with learning. This work is now published in Journal of Memory and Language.

Second, we have begun to unpack the various contributors to school-readiness in terms of preparedness for literacy learning. In computational models we have distinguished exposure, vocabulary size, and aspects of speed of processing and learning during a child's pre-literate experience with language. We have tested these various effects on the child's development of literacy. We have found that breadth as well as experience of vocabulary before reading are critical factors in reading success, and submitted this work for journal publication. We have also shown that training on phonics is only effective if the child has established good comprehension skills prior to beginning reading training. These are new and exciting findings, with great potential for broad academic and societal impact. We have published a 6-page conference paper providing the first report of these effects.

Third, the model of reading raised a prediction that words learned before literacy and words learning after literacy would be processed distinctively in the mind's reading neural network - with pre-literacy words engaging more the meaning of words than post-literacy words. We have shown that the model's predictions are exhibited in behavioural data on naming and lexical decision responses. A first report of this finding is published as a 6-page conference paper.
Exploitation Route We have implemented the first realistically trained model of reading. This will be made available to other interested researchers who wish to determine the effect of gradual learning on literacy development.
We have isolated the role of different contributors to literacy success, these can be taken forward by further testing of the predictions of the computational model, and design of interventions for improving literacy development in young children.
We have identified long-lasting differences in representation of words learned before and after literacy, which has profound implications for the resilience of different aspects of the vocabulary during aging.
Sectors Education

 
Description A consequence of our research has been identifying the role of precursors of literacy and testing these in computational models. For instance, previous behavioural studies have established correlations between vocabulary size, speed of processing, and language exposure on development of literacy. However, in computational models the causal contributions of each of these can be assessed, and teased apart. Their relative contributions to different aspects of reading research (e.g., phonological fluency, comprehension) can also be appraised. A key research finding of our computational work has been that children's pre-literacy oral vocabulary is critically important for effective literacy development, and has a profound influence on the effectiveness of different reading training programmes - phonics focused or word-focused training. We have been conveying our program of research and early results to The National Literacy Trust and The Communication Trust, two charities involved in development of language skills and with key interest in school-readiness, and to research partners in the Royal College of Speech and Language Therapists. Key events we have presented at are the National Literacy Trust's annual conference for early-years practitioners (2018) and at events for preschool practitioners as part of the Nursery World Forum (2017, 2018). We have also been engaging with caregivers of young children regarding the importance of shared book reading and language development skills prior to learning to read formally in school. We have emphasised the importance of variation as well as exposure to caregivers, and feedback has been positive in using a variety of books to ensure variation in exposure for children. We aim to provide impact in changing caregivers' behaviour in shared book reading, once our publications have been peer-reviewed we will work with library services and communicative development charities to highlight our message.
First Year Of Impact 2016
Sector Education
Impact Types Societal

 
Description Brysbaert Frequency lifetime 
Organisation University of Ghent
Department Faculty of Psychology and Educational Sciences
Country Belgium 
Sector Academic/University 
PI Contribution Invited by Professor Marc Brysbaert, we extended our computational models of reading to address how psycholinguistic factors might change in their influence on reading over the lifespan. We conducted computational models of monolingual and bilingual readers over the lifespan.
Collaborator Contribution Professor Brysbaert contributed the theory and toward the writing of the first report of the research.
Impact submitted paper: Monaghan, P., Chang, Y.N., Welbourne, S., & Brysbaert, M. (submitted). Exploring the relations between word frequency, language exposure, and bilingualism in a computational model of reading. Journal of Memory and Language, submitted. Multidisciplinary: computer science, psychology, linguistics.
Start Year 2015
 
Description Seidenberg - precursors of literacy 
Organisation University of Wisconsin-Madison
Country United States 
Sector Academic/University 
PI Contribution We are hosting visit of Prof Mark Seidenberg from University of Wisconsin at Madison - a foremost authority on computational modelling of literacy - in June/July 2016 to continue our collaborations which have commenced by correspondence. With Prof Seidenberg we will work on developing models of precursors of literacy, determining how language skills and exposure cause, or correlate with, literacy development.
Collaborator Contribution Prof Seidenberg will visit Lancaster University for four weeks to continue this collaboration.
Impact No outputs yet. Collaboration is multidisciplinary, combining computer science, linguistics, psychology, and educational sciences.
Start Year 2016
 
Description Parent engagement over book reading 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach Local
Primary Audience Patients, carers and/or patient groups
Results and Impact 623 children and caregivers visited a pop-up shop in Lancaster city centre ("Campus in the City") where we discussed with parents their book reading with children, determining how children engage with books across a variety of contexts, and parental attitudes to books and reading. Feedback from the event was extremely positive, with recruitment of 72 children to our databases for experimental studies, and positive responses from children and caregivers for some of the "prizes" of books that we sent out after the event.
Year(s) Of Engagement Activity 2016
 
Description Talk at conference NCPW14 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Other audiences
Results and Impact Talk to 80 academic international delegates on effects of comprehension and phonological deficits on reading acquisition in a computational model, followed up with requests for further information and questions and discussion.
Year(s) Of Engagement Activity 2014
URL http://www.psych.lancs.ac.uk/ncpw14/
 
Description Talk at conference UK orthography group 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Other audiences
Results and Impact Talk by Ya-ning Chang at UK Orthography Group meeting, attended by academics from UK, on age of acquisition effects in early reading.
Year(s) Of Engagement Activity 2015
URL http://www.bristol.ac.uk/expsych/events/2015/253.html