Application to Automatic Language Learning Sample Clauses

Application to Automatic Language Learning. Since the 1990s, there has been a sea change towards the use of statistical, corpus-based methods in all areas of computational linguistics, including the computational modeling of language acquisition. Work in this field constitutes a relatively recent addition to the methodological repertoire of developmental science (cf. Xxxxxxxxxx and Xxxxx 1997; Xxxxx 2006; Xxxxxx et al. 2008), and it has provided support for several important tenets of usage-based theories of language and its acquisition (cf. x.x. Xxxxx et al. 2005; Xxxxxxxxxxx et al. 2008; Xxxxxxxx and Xxxxxxxxx 2008). Also in the community of theoretical computational linguistics, which had traditionally seen the grammar learning problem to be intractable without Universal Grammar in view of Gold's results (Gold 1967), biases in the data such as typically found in CDS are beginning to be recognized as factors that ameliorate learning difficulty (Xxxxxxxx 2001; Xxxxx 2004; Xxxxx 2006). However, the algorithms which such approaches use to distil grammars from corpora are usually not only semantically blind, but also provided with certain grammatical information from the outset (e.g. part-of-speech annotation). From a developmental perspective, neither of these two features carries over to human learners – children ground linguistic signs in embodied experience, and they are not assumed to be equipped with adult syntactic categories such as ‘preposition’ or ‘conjunction’ from birth. Moreover, early caretaker-child interaction is restricted to joint attention scenarios (Xxxxxxx and Xxxxxx 2004), which is a further property that lacks in these approaches.