Many recent models of language comprehension have stressed the role of distributional frequencies in determining the relative accessibility or ease of processing associated with a particular lexical item or sentence structure. However, there exist relatively few comprehensive analyses of structural frequencies, and little consideration has been given to the appro- priateness of using any particular set of corpus frequencies in modeling human language. We provide a comprehensive set of structural frequencies for a variety of written and spoken corpora, focusing on structures that have played a critical role in debates on normal psycholinguistics, aphasia, and child language acquisition, and compare our results with those from several recent papers to illustrate the implications and limitations of using corpus data in psycholinguistic research.

It has previously been assumed in the psycholinguistic literature that finite-state models of language are crucially limited in their explanatory power by the locality of the probability distribution and the narrow scope of information used by the model. We show that a simple computational model (a bigram part-of-speech tagger based on the design used by Corley and Crocker (2000)) makes correct predictions on processing difficulty observed in a wide range of empirical sentence processing data. We use two modes of evaluation: one that relies on comparison with a control sentence, paralleling practice in human studies; another that measures probability drop in the disambiguating region of the sentence. Both are surprisingly good indicators of the processing difficulty of garden-path sentences. The sentences tested are drawn from published sources and systematically explore five different types of ambiguity: previous studies have been narrower in scope and smaller in scale. We do not deny the limitations of finite-state models, but argue that our results show that their usefulness has been underestimated. 1

That comprehension of object relatives constitutes a difficult domain in acquisition and pathology as well as in adult parsing, is a well known and robustly established fact (see Friedmann, Belletti, Rizzi (2009) for relevant references on classical work on acquisition and parsing; Adani (2008) for recent results on acquisition; Grillo (2008)

A controversy in the sentence processing literature regarding relative-clause processing was raised by Hsiao and Gibson’s (2003) study of Chinese relative clauses. Their study suggested that, contrary to

Abstract not found

This is the pre-publication manuscript. The published version may slightly differ.

Abstract not found

The good news for advocates of Construction Grammar (CxG) is that language scholars from a wide array of backgrounds have adopted its fundamental insight: knowledge of language includes grammatical generalizations of varied grains. CxG, or constructionbased syntax more generally, informs models of acquisition (Tomasello 2003, Diessel

Relative clauses containing subject relative-pronouns are the prevalent type both across languages and in conversation, accounting for 65 % of relative clauses in the American National Corpus (Reali and Christiansen 2007). This fact appears attributable to processing constraints, as per Hawkins 1999, 2004: subject extractions are the most local filler-gap dependency and therefore impose the least burden on short-term memory. This processing explanation, however, not only lacks strong psycholinguistic support but also fails to explain a major pattern in English corpora: subject relatives are the preferred modifiers only of object nominals. We propose that this preference is driven not by general-purpose interpretive or encoding constraints but by constructional licensing: the subject relative is part of an entrenched syntactic routine, the Presentational Relative construction (McCawley 1981, Lambrecht 1987, 1988, 2002). We investigate this hypothesis by examining the formal, semantic and pragmatic properties of relative-clause modifiers of object and oblique nominals in an English conversational corpus. We find that the subject relative-clauses within this set display significantly more hallmarks of presentational function than do their nonsubject counterparts. We conclude accordingly that the prevalence of subject relatives is a reflex of the role that this pattern plays in a frequently used construction, rather than constraints that rule out other relative-clause patterns.

In morphology, researchers have provided compelling evidence for the storage of even fully compositional structures that could otherwise be computed by rule. For example, a high-frequency word composed of multiple morphemes (e.g., root + plural inflection) may be stored directly rather than computed on the fly (e.g., Baayen, Dijkstra, & Schreuder, 1997). Here, we investigate similar questions of storage and computation in syntax, a domain in which evidence of storage of fully compositional structures has been less forthcoming. We approach this question using syntactic priming, a method exploiting the tendency of individuals to repeat syntactic structures that they have recently produced (Bock, 1986). As a test case, we investigate relative clauses (RCs). RCs are both abstract and syntactically complex but are nevertheless frequent in natural language (Roland, Dick, & Elman, 2007). Moreover, differences in processing complexity between different RC types are at least partially predicted by frequency (e.g., Reali & Christiansen, 2007). RCs are therefore an ideal domain to look for evidence of storage of abstract, compositional syntactic structure. If the structures underlying high-frequency RC types are stored and retrieved from memory as whole units instead of being computed online from smaller units, then these stored structures should be susceptible to priming. Across three experiments, we observed that priming of object-extracted RCs is sensitive to a) the type of noun phrase in the embedded subject position (a full NP vs. a pronoun), and b) the type of relative pronoun (who vs. that). These results suggest that the representations of some types of RCs involve storage of large units which include both syntactic and lexical information. We interpret these results as providing support for models of syntax that allow for complex mixtures of storage and computation.