Verb bias and structural priming in non-linguistic grammar acquisition task
2013; Wiley; Volume: 35; Issue: 35 Linguagem: Inglês
ISSN
1551-6709
AutoresMarius Janciauskas, Franklin Chang,
Tópico(s)Reading and Literacy Development
ResumoVerb bias and structural priming in non-linguistic grammar acquisition task Marius Janciauskas (m.janciauskas@liv.ac.uk) Franklin Chang (franklin.chang@liverpool.ac.uk) University of Liverpool, School of Psychology, Bedford Street South, Liverpool, L69 7ZA UK Abstract said to have double object dative bias. This phenomenon is thought to occur as a result of learning distributional relationships between verbs and structures (Juliano & Tanenhaus, 1994). In this example, the DO bias arises from stronger probabilistic association of the verb ‘give’ with the DO structure. A verb’s occurrence in its preferred structure (verb-structure match hereafter) is known to influence structural choices and reduce processing time at the choice point where alternating structures diverge (Ferreira, 1996; Garnsey, Pearlmutter, Myers, & Lotocky, 1997; Stallings, MacDonald, & O’Seaghdha, 1998). Another phenomenon of interest is structural priming, which is the tendency for participants to repeat previously produced sentence structures (Bock, 1986). For example, if participants heard the DO sentence like “the boy threw the dog the ball” and are then given a picture which can be described using a DO (e.g. “the man gave the woman the dress”) or a PD (e.g. “the man gave the dress to the woman”) structure sentence, they were more likely to use the same DO structure. Structural priming has been found to persist over time, suggesting that it is supported by learning (Bock & Griffin, 2000). Chang, Dell and Bock (2006) used a connectionist model to show that priming could be explained as SL over abstract structural representations. Like verb bias, structural priming influences structural choices in sentence production and comprehension times at the post-verbal position (Corley & Scheepers, 2002; Weber & Indefrey, 2009). In sum, verb bias and structural priming are thought to depend on SL processes involving linguistic units. If these processes are not specific to language then it should be possible to find verb bias- and structural priming-like effects in a non-linguistic SRT task. The present studies are a step towards such a paradigm. Domain-general statistical learning (SL) is thought to support language phenomena like verb bias and structural priming. We explored this idea by inducing these phenomena within a non-linguistic serial reaction time (SRT) task where participants learned an English-like artificial language using SL. In a series of two experiments we found error rates to be sensitive to verbs’ structural preferences and abstract structural priming. The similarities between the behaviour in this task and previous linguistic research suggests that this method may be useful for studying the nature of SL in language learning and processing. Keywords: statistical learning; verb bias; structural priming. An important question in the study of language is the degree to which language acquisition depends on language-specific mechanisms or general-purpose statistical learning (SL) mechanisms (e.g., Kidd, 2012). Research has found that SL takes place in real and artificial language learning tasks (Fine & Jaeger, 2013; Qi, 2012; Saffran, 2003, Wells, Christiansen, Race, Acheson, & MacDonald, 2009). However, the use of language or auditory stimuli could cause language-specific systems to become activated in these tasks (Gervain, Nespor, Mazuka, Horie, & Mehler, 2008). Non-linguistic artificial grammar learning (AGL) or serial reaction time (SRT) tasks provide a paradigm for studying grammar learning that is independent of linguistic knowledge. But the grammars used in the existing studies (e.g. Hunt & Aslin, 2010) are quite different from real language and it is hard to link findings in these studies to human syntactic phenomena. Thus, it is still not known if domain-general SL can account for the acquisition and processing of human syntactic knowledge. The present study set out to develop a method to study SL processes within a non-linguistic task designed to approximate the contexts in which certain linguistic phenomena occur. We developed an SRT task where participants had to implicitly learn statistical regularities in symbol sequences generated from an English-like grammar in a symbol-matching task. If participants learn this language as they process it (linguistic adaptation; Chang, Janciauskas, & Fitz, 2012), then their accuracy and reaction times should reveal how linguistic phenomena arise out of general-purpose SL. We applied our paradigm to explain two language phenomena: verb bias and structural priming. Verb bias is the tendency for individual verbs to prefer particular structures. For example, if a verb occurs more often in the double object (DO) structure as in “the man gave the woman the dress” rather than the prepositional dative (PD) structure “the man gave the dress to the woman”, the verb is Study 1: Dative Alternation SRT Task The first study used a variant of Hunt and Aslin’s (2001) SRT study. In the centre of a computer screen participants saw sequences of letters appearing one at a time, which required them to find that letter on a circle of 21 letters surrounding the centre by moving a mouse cursor over it. The sequences were structured based on a grammar that included English dative alternation-like structures. For example, the symbol string “H J Z C M” approximated a PD sentence without articles like “man gave dress to woman”. The corresponding DO symbol string was “H J M Z” (“man gave woman dress”). Verb bias was created by varying the frequency of the symbols (verbs hereafter) appearing in the verb’s position with particular structures. For example, J and B occurred more often with PD structure, while D and
Referência(s)