Empiricist Solutions to Nativist Puzzles by means of Unsupervised TSG

While the debate between nativism and empiricism exists since several decades, surprisingly few common learning problems have been proposed for assessing the two opposing views. Most empiricist researchers have focused on a relatively small number of linguistic problems, such as Auxiliary Fronting or Anaphoric One. In the current paper we extend the number of common test cases to a much larger series of problems related to wh-questions, relative clause formation, topicalization, extraposition from NP and left dislocation. We show that these hard cases can be empirically solved by an unsupervised tree-substitution grammar inferred from child-directed input in the Adam corpus (Childes database). 1 Nativism versus Empiricism How much knowledge of language is innate and how much is learned through experience? The nativist view endorses that there is an innate language-specific component and that human language acquisition is guided by innate rules and constraints (“Universal Grammar”). The empiricist view assumes that there is no language-specific component and that language acquisition is the product of abstractions from empirical input by means of general cognitive capabilities. Despite the apparent opposition between these two views, the essence of the debate lies often in the relative contribution of prior knowledge and linguistic experience (cf. Lidz et al. 2003; Clark and Lappin 2011; Ambridge & Lieven 2011). Following the nativist view, the linguistic evidence is so hopelessly underdetermined that innate components are necessary. This Argument from the Poverty of the Stimulus can be phrased as follows (see Pullum & Scholz 2002 for a detailed discussion): (i) Children acquire a certain linguistic phenomenon (ii) The linguistic input does not give enough evidence for acquiring the phenomenon (iii) There has to be an innate component for the phenomenon In this paper we will falsify step (ii) for a large number of linguistic phenomena that have been considered “parade cases” of innate constraints (Crain 1991; Adger 2003; Crain and Thornton 2006). We will show that even if a linguistic phenomenon is not in a child‟s input, it can be learned by an „ideal‟ learner from a tiny fraction of childdirected utterances, namely by combining fragments from these utterances using the Adam corpus in the Childes database (MacWhinney 2000). Previous work on empirically solving nativist puzzles, focused on a relatively small set of phenomena such as auxiliary fronting (Reali & Christiansen 2005; Clark and Eyraud 2006) and Anaphoric One (Foraker et al. 2009). Some of the proposed solutions were based on linear models, such as trigram models (Reali & Christiansen 2005), though Kam et al. (2008) showed that the success of these models depend on accidental English facts. Other empiricist approaches have taken the notion of structural dependency together with a