Evolving Connectionist Models to Capture Population Variability Across Language Development: Modelling Children’s Past Tense Formation

Children’s acquisition of English past tense has been widely studied as testing ground for theories of language development, mostly because it comprises a set of quasi-regular mappings. English verbs are of two types: regular verbs, which form their past tense based on a productive rule, and irregular verbs, which form their past tenses through exceptions to that rule. Although many connectionist models exist for capturing language development, very few consider individual differences. In this paper, we explore the use of populations of artificial neural networks (ANNs) that evolve according to Behavioural Genetics principles in order to create computational models capable of capturing the population variability exhibited by children in acquiring English past tense verbs. Literature in the field of Behavioural Genetics views variability in children’s learning in terms of genetic and environmental influences. In our model, the effects of genetic influences are simulated through variations in the neuro-computational properties of ANNs, and the effects of environmental influences are simulated via a filter applied to the training set. This filter alters the quality of information available to the artificial learning system and creates a unique subsample of the training set for each simulated individual. Our approach uses a population of twins to disentangle genetic and environmental influences on past tense performance and to capture the wide range of variability exhibited by children as they learn English past tenses. We use a novel technique to create the population of artificial neural network twins based on the biological processes of meiosis and fertilization. This approach allows modelling both individual differences and development (within the lifespan of an individual) in a single framework. Finally, our approach permits the application of Selection on developmental performance on the quasi-regular task across generations. This is an important aspect that distinguishes our work from others reported in literature, setting individual differences within an evolutionary framework. We present an experimental evaluation of this model focusing on individual differences in performance. The experiments led to some interesting findings such as: applying selection on the individual’s performance level in a quasi-regular task such as past tense acquisition results in the emergence of divergent behaviours depending on initial conditions – both genetic and environmental; once selection starts targeting a particular aspect of task domain, it starts behaving similar to Waddington’s epigenetic landscape; and selection based on a stochastic method such as roulette-wheel, when combined with sexual reproduction method for population generation, has a limiting effect on final behavioural (or performance) levels achieved. The findings validate the effectiveness of the method within an evolutionary setting and provide the basis for future work to capture population-level differences within a developmental setting. Maitrei Kohli *+ Birkbeck, University of London George D. Magoulas * Birkbeck, University of London Michael S.C. Thomas ** Birkbeck, University of London