Professor Julian Pine BA, PhD

Research Overview

My research focuses on how children learn words, how children learn to combine words into sentences and how these processes can be supported by caregivers and early years practitioners. It uses a range of methods, including corpus analysis, behavioural experiments and computational modelling and has a strong cross-linguistic focus. It also includes work on children with language difficulties.

Example Project: Modelling the cross-linguistic pattern of verb-marking deficits in children with Developmental Language Disorder

Developmental Language Disorder (DLD) refers to ‘a significant deficit in language ability that cannot be attributed to hearing loss, low non-verbal intelligence or neurological damage’ (Leonard, 2014: 3). Research suggests that approximately 7% of the preschool-age population exhibit this kind of developmental profile, with boys more likely to be affected than girls - 8% versus 6%, respectively (Tomblin et al, 1997).

Children with DLD are not a homogeneous population, and may show deficits in a number of different ways, including phonology, word learning, morpho-syntax and pragmatics. However, they tend to show a particular deficit in the acquisition and use of verb morphology. Interestingly, however, the size and nature of this deficit varies across languages. For example, children with DLD learning English show large deficits in verb-marking relative to both age-matched and language-matched controls (Rice, Wexler & Hershberger, 1998); children with DLD learning German show large deficits in verb marking relative to age-matched controls, but only marginal deficits relative to language-matched controls (Rice, Noll & Grimm, 1997); and, children with DLD learning Spanish show only marginal deficits in verb marking relative to age-matched controls and no deficits relative to language-matched controls (Bedore & Leonard, 2005).

In this project, we are using corpus analysis and computational modelling techniques to investigate the nature of the deficit that underlies the cross-linguistic pattern of verb marking error in children with DLD. This will be done by modifying an existing computational model of language learning (MOSAIC), which has already been shown to simulate the pattern of verb marking error in typically developing children learning English, German and Spanish (Freudenthal, Pine, Aguado-Orea & Gobet, 2007; Freudenthal, Pine, Jones & Gobet, 2015). Additional mechanisms will be implemented in MOSAIC based on processes that have been argued to underlie the pattern of verb marking error in typically developing children and children with DLD learning English. These will include: 1) Defaulting on the basis of the relative frequency of different forms of the verb (Räsänen, Ambridge & Pine, 2014; Kueser, Leonard & Deevy, 2018); 2) Learning from inappropriate sentential contexts (Theakston, Lieven & Tomasello, 2003; Leonard, Fey, Deevy & Bredin-Oja, 2015) and 3) Producing telegraphic speech (Conti-Ramsden, Botting & Farragher, 2001) as a result of limited sequence learning ability (Hsu, Tomblin & Christiansen, 2014)

Models incorporating each of these mechanisms will be used to investigate a) the extent to which incorporating each mechanism improves MOSAIC’s fit to the cross-linguistic data on TD children, and b) the extent to which reducing the rate at which the effect of each mechanism decreases over development allows us to capture cross-linguistic variation in the pattern of verb marking deficit in children with DLD. Of particular interest will be whether it is possible to use exactly the same version of the model to simulate the large deficit in verb-marking in children with DLD learning English relative to language-matched controls, the much smaller deficit in verb marking in children with DLD learning German relative to language-matched controls, and the lack of any deficit in verb-marking in children with DLD learning Spanish relative to language-matched controls.