Language Abstraction: Consolidation of Language Structure During Sleep

Language Abstraction: Consolidation of Language Structure During Sleep Michelle C. St. Clair (michelle.stclair@manchester.ac.uk) School of Psychological Sciences, University of Manchester Manchester, M13 9PL, UK Padraic Monaghan (p.monaghan@lancaster.ac.uk) Department of Psychology, University of Lancaster Lancaster, LA1 4YF, UK However, before positing specific language-learning mechanisms or innately-specified linguistic structure, it is important to do full justice to the potential richness of general purpose learning mechanisms for the acquisition of language structure. There are many candidates for studies showing the great potential of linguistic input (Monaghan, Christiansen, & Chater, 2007) and the versatility of general learning mechanisms that can generate structure from the language itself (e.g., Reali, Christiansen, & Monaghan, 2003). In the latter study, a simple recurrent network was trained to learn the patterns of associations between distributional and phonological cues for categorising grammatical categories in speech. So, a fundamental question for language acquisition research is what are the range of general purpose learning mechanisms available to the language learner, and how do they apply to language input? There is accumulating evidence that there are critical qualitative distinctions in the effects of learning new material when assessed before compared to after sleep, and specifically that sleep may facilitate abstraction and generalisation in a range of tasks (e.g., Fischer, Drosopoulos, Tsen, & Born, 2006; Gomez, Bootzin, & Nadel, 2006; Wagner, Gais, Haider, Verleger, & Born, 2004). Although only a few sleep studies have used linguistic tasks, the underlying structures and features of the stimuli used in many of these studies are strikingly similar to the critical features of natural and artificial languages. The current study advances the established field of language learning studies by making use of innovative recent methods for investigating the cognitive influences of sleep on complex learning. We contend that sleep elicits critical general purpose learning that is important for the process of language acquisition. Abstract Recent research on how sleep influences cognitive processes has pointed to the potential role of sleep in the consolidation and generalisation of information across a range of tasks, including some language learning tasks (i.e., Gomez, Bootzin, & Nadel, 2006; Wagner, Gais, Haider, Verleger, & Born, 2004; Walker, 2005). The current study investigated the role of sleep in grammatical category generalisation by testing participants for categorical knowledge 12 hours after initial training. Participants were tested on their categorisation of the training words and also on their ability to generalize to novel, but consistent category words. The sleep group received training at 9pm and the wake group at 9am. Only the sleep group was able to abstract the general category characteristics to the novel category words which implies a domain general abstraction mechanism may be influential in language acquisition. Introduction The child’s language environment provides an immensely rich source of information that can contribute to the development of structure at the word level (Saffran, Aslin, & Newport, 1996) and to the development of grammatical categories (Redington, Chater, & Finch, 1998) and syntactic constraints (Bod, 2007). From a sea of sounds (Saffran, 2001), the child must learn where words begin and end in continuous speech, and the child must also learn the constraints on the relations between words, as described by grammatical structure. One of the central issues in the cognitive science of language over the last 50 years is to determine the extent to which the language environment triggers innately-specified, domain-specific processing (Chomsky, 2006; Pinker, 1991), and the extent to which sufficient information is present within the signal to provide structure via general-purpose learning mechanisms (Conway & Christiansen, 2005; Saffran, 2002). Chomsky (1965, 1980) proposed that the linguistic input that a child is exposed to provides insufficient information to adequately constrain the actual grammar of the language that the child develops (Gold, 1967). For instance, children are unlikely to be exposed to instances of auxiliary-fronting (e.g., converting “the dog that is sleeping is grey” to a question format: “is the dog that is sleeping grey” by moving the correct auxiliary verb). The debate over whether general-purpose learning mechanisms can provide sufficient, indirect constraints to distinguish between grammatical versus ungrammatical constructions is a matter of current contention (see, e.g., Messer et al., 1995; Reali & Christiansen, 2005). Sleep Consolidation and Qualitative Changes in Learning Sleep has been found to play a pivotal role in memory consolidation, both in terms of declarative and procedural memory, as well as being implicated in brain plasticity for storage (see Walker, 2005 for a review). Increases in perceptual skills and in explicit knowledge about complex tasks has also been associated with sleep (Fischer, Drosopoulos, Tsen, & Born, 2006; Wagner, Gais, Haider, Verleger, & Born, 2004). Wagner et al. (2004) conducted a study to test learning of abstract structure in strings of digits, where the last three digits were the second, third and fourth digits in reverse order (e.g. 1914419). After an eight hour