While psycholinguistic studies show that sentence processing goes incrementally "from left to right" (i.e. sequentially with respect to how words are produced and perceived, Bever 1970, Fodor 1978, De Vincenzi 1990, MacDonald et al. 1994, a.o.), the prevailing models of theoretical syntax are often indifferent (Pollard and Sag 1994) or even in conflict (Chomsky 1995) with this assumption of incrementality and lexical items processing order. A serious attempt to fill this interdisciplinary gap was started by Phillips (1996) that reshaped the basic phrase structure building operations so as to allow incremental constructions of syntactic constituents from left to right; in addition, the implementation of Parsing algorithms (Kobele 2012) and complex NLP systems (e.g. a machine translation system from verbal Italian to Italian Sign Language, Chesi et al., 2008, Geraci & Chesi 2009) has been possible only by virtue of fully formalized, and easily upgradable, minimalist grammars support (Stabler 1997, Chesi 2004, 2012).

 

On the one hand, from a formal point of view, this approach showed that sequential processing ("left-right") is empirically advantageous (Phillips 2002), on the other, that the orientation of the syntactic derivation should be logically re-orientated top-down (especially in terms of lexical insertion and Filler-Gap dependency, Fodor 1978) (Chesi 2004-2012). This plainly allowed to explain a number of otherwise intricate empirical data, traditionally investigated by theoretical approaches (parasitic gap constructions, quantifiers scope, restrictions on filler-gap dependencies and rightward movement, cf. Bianchi & Chesi 2006, 2008, 2012, Chesi 2013 respectively) and also experimental data (results of self-paced reading, eye-tracking and neuroimaging experiments, cf. Mancini et al. 2013, on filler-gap dependencies, Sprout et al. 2013). Recently, the simplification of the basic grammatical operations (Chomsky 2005) reduced the complexity of the theory and increased its explanatory power, raising fundamental questions about the neurophysiology of language and the acquisition and processing of grammatical rules (Moro et al. 2001, Hood, 2011), although it did not completely solve this tension.

On the other hand, in terms of Natural Language Processing, it has been shown that the explicit formalization of linguistic competence allows us to implement more efficient Parsers, where purely statistical models productively integrate syntactic restrictions (Collins 1999).

 At present, therefore, the integration of formal-linguistic models with neuro/psycho-linguistic and computational ones becomes a priority.

The urgency of investigating this (that is the main goal of this project) is demonstrated by the main topic of the recent 25th Annual CUNY Conference on Human Sentence Processing, organized by J. Fodor (New York, 14-16/03/2012, Sponsor: National Science Foundation): discussing more promising models of Linguistic Competence and (psycholinguistic/computational) Performance. At this conference, Chesi was Invited Speaker (presenting the formalism discussed in this project), together with N. Chomsky, S. Fong, L. Frazier, C. Phillips, I. Sag and M. Steedman (http://cuny2012.commons.gc.cuny.edu/).