THEORETICAL FRAMEWORK OF THE PROJECT AND A DESCRIPTION OF THE STATE-OF-THE-ART
Since Chomsky’s Syntactic Structures (1957), the notion of mental grammar, traditionally seen as the set of principles of combination of language, has been confronted with the finite set of memorized units that are traditionally called the lexicon. As Jackendoff (1997) states, an alternative to that is to apply the term mental grammar more broadly, so that it includes both the lexicon and the computational system. Although one can understand the reasons for the principal distinction between lexicon and grammar, the development of the field in the last twenty-five years of the 20th century brought to our attention the domains of their overlap and unification (Pustejovsky 1991; Bates et al. 1997; Schönefeld 2001; Haspelmath 2007; Hilpert 2008. etc.).
Among the main lines of research that challenged this divide are contemporary psycholinguistic processing studies (MacDonald 1994; Bates et al. 1997a; Bock 1999; Levelt et al. 1999; Ferreira and Dell 2000; Wagner et al. 2010; Viglioco et al. 2014. etc.). They brought under scrutiny the assumption of a strict divide between lexicon and grammar, or the partition of knowledge between the lexical list (or, to express it in a more complex fashion, semantic net) and the rules of grammar that cannot be justified by analyzing linguistic phenomena in natural languages.
The notion of mental grammar, understood as a complete collection of patterns, templates or schemas of the language, stored in the brain of a language user, represents the central theoretical construct of modern linguistics. Since we do not have direct access to mental grammar (Jackendoff 2012 (2016)), we choose to investigate its relationship to the constraints of information structure, primarily at the interface of syntax and semantics. We understand information structure as a component of sentence grammar as it was proposed by Lambrecht (1994). In this view, information structure is seen as a decisive factor in formal structuring of a sentence. The link of mental grammar to information structure is grounded in the fact that “making sense” involves, among other things, knowledge of patterns. This is why we must ascribe to the speaker’s mind a mental grammar that specifies possible sentence patterns.
Both production and comprehension studies prove to be a reliable testing ground for the relationship of lexicon and syntax, especially for their mutual interdependence in a much more substantial way than the one anticipated by traditional generative grammar. Both syntactically and semantically driven models of language processing are in line with the psychological reality that every lexeme sets the frame for its place in the language system. The frame has to be defined in terms of conceptual structure, as well as in terms of the inherent morphosyntactic features that place constraints on the licensing domain of every particular lexeme. After introducing a particular lexeme in a sentence, only a limited set of other lexemes and grammatical forms can be selected (See more in Schönefeld 2001, ch. 3 and 4.).
Although classic generative approaches build on the differentiation between lexicon and grammar, there are many theories (Cognitive Grammar, Langacker 1987; Emergent Grammar, Hopper 1987; Construction Grammar, Goldberg 1995, 2006; Parallel Architecture, Jackendoff 2002), particular insights (Schönefeld 2001; Haspelmath 2007, 2014), as well as experimental approaches (Bates and Goodman 1997; Hilpert 2008) that advocate the inseparability of grammar and lexicon. All of them depart from the principal distinction between the rules and the lexemes. The rules are no more treated as procedures, but as templates or schemas (pieces of structure with variables), which brings them closer to lexemes. As a theoretical baseline we will adopt this, fundamentally constructivist theoretical approach, using Jackendoff’s Parallel Architecture (PA) (Jackendoff 1983, 1987, 1990, 1997, 2002) where appropriate.
In PA there is no strict division between lexicon and grammar. Rather, words and standard rules are at the opposite corners of a multidimensional continuum that includes all sorts of mixed items such as idioms and meaningful constructions. An important consequence of this view is that semantics in not necessarily in a one-to-one relation to syntax, but most often in a many-to-many relation, which needs to be explicitly stated. What this means is that at the sentence level, phonology, semantics and syntax are not derived from each other. Their mutual relations are established through interface rules, which are in some cases exhaustive and in others subject to variability. At word level, every lexical item is a triple consisting of small pieces of phonological, syntactic and semantic structure (Jackendoff, 2002). This makes words prototypical interface rules.
Based on broadly understood agreement potential, in this project we will explore the applicability of the PA as a theoretical basis for determining the set of necessary categorical relations within the lexicon, both at the morphosyntactic, and the semantic level. In order to track down the features that need to be specified in Croatian as a highly inflected language, we will first model a description of representational levels and interfaces between them for abstract and concrete and high and low imageable nouns, verbs and adjectives, especially taking into account specificities of their agreement potential. Our analysis distinguishes between the need for specification of universal and language-specific features.
Existing descriptions of Croatian largely keep grammar and lexicon divided. There is a single recent book-length attempt of describing Croatian within the framework of Cognitive Grammar (Belaj 2015) which does not make the grammar-lexicon distinction, and there is a number of recent papers dealing with a variety of issues within Croatian grammar from this point of view (e.g. Stanojević 2011). Importantly, none of these approaches combine a corpus-based study with psycholinguistic experiments. Recently, there has been a single group of researchers working within a similar vein on Slavic modality (led by Dagmar Divjak, and including Peti-Stantić and Stanojević), but their publications have thus far not included Croatian. Therefore, the presentation of nominal and verbal representational geometry that takes into account levels and interfaces that are already specified in the lexical form is a novel approach. The comparison of number and the hierarchy of features needed to be specified for languages of distinct morphosyntactic richness (i.e. English and Croatian) enables us to establish a hierarchy of features themselves. Our proposal is to use this hierarchy as a baseline for experimental work on variability and language processing.