According to Chomsky (2000b: 119), ‘the human language faculty and the (I–)languages that are manifestations of it qualify as natural objects’. This approach – which ‘regards the language faculty as an “organ of the body”’ – has been labelled the ‘biolinguistic perspective’ by Chomsky (2005: 1). Hauser, Chomsky and Fitch (2002: 1570) base their discussion of the key biological question of evolution on the ‘biologically and individually grounded’ use of the term language ‘to refer to an internal component of the mind/brain (sometimes called “internal language” or “I-language”)’. They distinguish two conceptions of the faculty of language, one broader (FLB) and one narrower (FLN):
FLB includes FLN combined with at least two other organism-internal systems, which we call ‘sensory-motor’ and ‘conceptual-intentional’ … A key component of FLN is a computational system (narrow syntax) that generates internal representations and maps them into the sensory-motor interface by the phonological system and into the conceptual-intentional interface by the (formal) semantics system … Most, if not all, of FLB is based on mechanisms shared with nonhuman animals … FLN – the computational mechanism of recursion – is recently evolved and unique to our species. (Hauser, Chomsky and Fitch 2002: 1571)
The conception of the language faculty and of (I-)languages as ‘natural’, ‘biologically grounded’ objects corresponds to specific theories concerning their internal articulation:
the I-language consists of a computational procedure and a lexicon. The lexicon is a collection of items, each a complex of properties (called ‘features’) … The computational procedure maps an array of lexical choices into a pair of symbolic objects, phonetic form and LF [logical form] … The elements of these symbolic objects can be called ‘phonetic’ and ‘semantic’ features, respectively, but we should bear in mind that all of this is pure syntax and completely internalist. (Chomsky 2000b: 120)
The internal articulation of the FLN is crucial to the biolinguistic programme, no less than its applications to domains such as language evolution, genetics and neurology. Here we address some points concerning this; specifically, we concentrate on the issue of language variation, starting with the idea that ‘the diversity and complexity can be no more than superficial appearance … the search for explanatory adequacy requires that language structure must be invariant’ (Chomsky 2000b: 7), and ‘There is a reason to believe that the computational component is invariant, virtually … language variation appears to reside in the lexicon’ (Chomsky 2000b: 120).
From this perspective, a central aim of our work is to provide empirical support for what we may call the lexical parametrization hypothesis (Manzini and Wexler 1987), and thus to make more precise the sense in which it holds. Without a doubt ‘one aspect is “Saussurean arbitrariness”, the arbitrary links between concepts and sounds … However, the possible sounds are narrowly constrained, and the concepts may be virtually fixed’ (Chomsky 2000b: 120). In the present study, we address the issue of how the linguistically relevant conceptual space yields different (I-)languages beyond the obvious aspect of ‘Saussurean arbitrariness’.
Before proceeding to the empirical core of the argument, we briefly introduce some of the conceptual underpinnings of the framework we adopt, beginning with the thesis that language ‘is a system that is, as far as we know, essentially uniform. Nobody has found any genetic differences … since its emergence there has not been any significant evolution. It has stayed that way’ (Chomsky 2002: 147). This view is shared by much current work on human cognitive and linguistic evolution (Lieberman 1991; Jackendoff 2002). The conclusion holds both for living languages and for ancient ones (whether documented and no longer spoken or merely reconstructed); as argued by Labov (1994), the same mechanisms of (surface) variation and change affect all of them. To take a comparative typological perspective:
no evidence of anything like speciation has been found … Languages from typologically very different areas have the same latent structural potential … this survey has uncovered no evidence that human language in general has changed since the earliest stage recoverable by the method used here. There is simply diversity, distributed geographically. (Nichols 1992: 227)
As for this geographically distributed diversity:
a residual zone or a set of residual zones will contain a good deal of the world’s possible linguistic diversity in microcosm, and both the existence of internal diversity and its actual profile are stable and obviously very natural situations. Diversity of a particular kind may even be regarded as the state to which a group of languages will naturally revert if left undisturbed … Spread zones, in contrast, are typically highly divergent from one another, but each is internally quite homogeneous … Just which language spreads in a spread zone is a matter of historical accident, and this historical accident can distort the statistical distribution of linguistic types in an area. (Nichols 1992: 23)
The set of languages considered in this work presents the kind of variation that we expect in natural languages in the absence of external constraints. Because of the political and cultural factors which, for centuries, have kept the Italian peninsula in conditions of great administrative and social fragmentation, dialectal differentiation in Italy has been preserved for longer (i.e. up to the present day) than in other areas of Western Europe, including Romance-speaking ones. Thus Italian varieties provide a rich and articulated picture of language variation that contrasts with that of other intensively studied varieties such as those of English. The view we take is that it is linguistic situations such as those in Britain, for example, that represent a somewhat misleading picture of variation, reflecting not only the internal shaping forces of language development, but also external mechanisms of social and political standardization. The variation seen in Albanian, including the major Gheg vs. Tosk divide in mainland Albania, and Arbëresh varieties of Southern Italy, has the same general character as that observed in Romance varieties. In the internalist (i.e. ‘biologically, individually grounded’) perspective that we adopt, variation between two or more varieties (linguistic communities) is in fact not qualitatively different from variation within the same variety (community), or even within the production of a single speaker. For example, to the extent that a speaker alternates between stylistic levels according to the situation of use, s/he will have a ‘bilingual’ competence of sorts – which, given the lexical parametrization hypothesis adopted here, can be accounted for as the co-existence of different lexicons with a single computational component (MacSwan 2000).
Suppose, then, that the lexicon is the locus of linguistic variation – in the form of a uniform (i.e. invariant) computational component, and of an invariant repertory of interface primitives, both phonological and conceptual. Non-trivial questions arise at this point: how can the lexicon vary on the basis of a universal inventory of properties (or ‘features’), and why does that variation in the lexicon result in variation in order, agreement, selection, and other relations that are computationally determined? These questions are amply debated in current linguistic theory. Our empirical discussion aims to support certain positions emerging from the debate, as opposed to others which are in principle equally possible.
In particular, the answer to the preceding questions is mediated for various scholars by the notion that there is a fundamental distinction between functional and non-functional elements. Thus, within the Distributed Morphology framework, Embick (2000:187) assumes a ‘distinction between the functional and lexical vocabularies of a language … functional categories merely instantiate sets of abstract syntacticosemantic features’, on which the derivational component operates. The actual phonological terminals corresponding to these abstract categories are inserted only after a level of morphological structure, where readjustment rules apply (Late Insertion). It is evident that the overall architecture of the grammar implied by this model is considerably more complex than one in which ‘the formal role of lexical items is not that they are “inserted” into syntactic derivations, but rather that they establish the correspondence of certain syntactic constituents with phonological and conceptual structures’ (Jackendoff 2002: 131).
Kayne’s (2006, 2008a) parametrization model, while avoiding recourse to Late Insertion, is close to Distributed Morphology in assuming that functional items correspond to a universal lexicon of sorts. Lexical and hence grammatical differences depend on whether the elements of this functional lexicon are overtly realized or ‘silent’. Interestingly, for Kayne (2006), even variation in the substantive lexicon can be reduced to variation in functional structure in the sense just defined, as can be seen in his construal of shallow as ‘LITTLE deep’, that is, essentially as the specialized lexicalization of deep in the context of the silent functional category ‘little’.
Manzini and Savoia (2005, 2007, 2008a) pursue a model under which, again, there is a unified conception of lexical variation – however, this is of the type traditionally associated with the substantive lexicon: there is a conceptual and grammatical space to be lexicalized and variation results from the distinct partitioning of that space. There is no fixed functional lexicon which varies along the axis of overt vs. covert realization – so-called functional space is just like all other conceptual space, and all lexical entries are overt. Thus, the distinction between functional (i.e. grammatical) contents and conceptual ones is an external one; as such it may very well be useless, and at worst it may obscure the real underlying linguistic generalizations.
Our conception of variation within the so-called functional lexicon is consistent with current conclusions regarding the conceptual space and the different ways in which it surfaces in natural languages. Fodor (1983) and Jackendoff (1994), among others, develop the Chomskyan theme that concepts, like other aspects of language, must have an innate basis – largely because of the poverty of stimulus argument. It has already been observed by Lenneberg (1967) that lexical items are the overt marks of a categorization process through which human beings carve out an ontological system from the perceptual continuum of the external world. This process of categorization is of course only indirectly connected with the objects of the external world. Jackendoff (1994: 195) notes that the lexical forms employed to express spatial location and motion (e.g. The messenger is in Istanbul; The messenger went from Paris to Istanbul; The gang kept the messenger in Istanbul) typically also express possession (e.g. The money is Fred’s; The inheritance finally went to Fred; Fred kept the money), the ascription of properties (e.g. The light is red; The light went from green to red; The cop kept the light red), etc.
This suggests that thought has a set of precise underlying patterns that are applied to pretty much any semantic field we can think about. Such an underlying ‘grain’ to thought is just the kind of thing we should expect as part of the Universal Grammar of concepts; it’s the basic machinery that permits complex thought to be formulated at all. (Jackendoff 1994: 197)
Dehaene, Izard, Pica and Spelke (2006) study geometrical concepts in an isolated group of Amazonian people whose language, Mundurukú, ‘has few words dedicated to arithmetical, geometrical, or spatial concepts’. They conclude that
geometrical knowledge arises in humans independently of instruction, experience with maps or measurement devices, or mastery of a sophisticated geometrical language … There is little doubt that geometrical knowledge can be substantially enriched by cultural inventions such as maps, mathematical tools, or the geometrical terms of language … however, the spontaneous understanding of geometrical concepts and maps by this remote human community provides evidence that core geometrical knowledge, like basic arithmetic is a universal constituent of the human mind. (Dehaene, Izard, Pica and Spelke 2006: 385, our italics)
In a similar vein, Hespos and Spelke (2004) study the acquisition of the conceptual distinction between ‘tight’ and ‘loose’ fit of one object to another in English-speaking children, which is not lexicalized in English, though it is in other languages like Korean. Their conclusion is that ‘like adult Korean speakers but unlike adult English speakers, these infants detected this distinction … Language learning therefore seems to develop by linking linguistic forms to universal, pre-existing representations of sound and meaning’ (Hespos and Spelke 2004: 453).
In short, the building blocks that are combined to make up the potentially infinite variety of human lexicons are innate. The lexicons of different languages are formed on this universal basis, covering slightly different extensions of it and in slightly different ways. The view we advocate here is simply that ways of representing the event, such as transitivity or voice (chapters 5–6), ways of connecting arguments to predicates (or to one another), such as cases (chapters 7–8), and more, are to be thought of as part of this general system. There is no separate functional lexicon – and no separate way of accounting for its variation. We started with the general Chomskyan biolinguistic, or internalist, picture of language, and of its basic components, both broadly and narrowly construed. Variation is crucial to establishing this model for the obvious reason that the uniformity thesis, as laid out above, requires a suitably restrictive account of observed cross-linguistic differences. But, even more fundamentally, the lexical parametrization hypothesis that we adopt means that questions of variation will inevitably bear on the form of the lexicon, as one of the crucial components of the I-language.
The other main component of the I-language is ‘the computational procedure’, which ‘maps an array of lexical choices into a pair of symbolic objects, phonetic form and LF’ (Chomsky 2000b, quoted above). As for the latter, Culicover and Jackendoff (2005: 6) aptly characterize a particularly popular conception of the relation of LF to the syntax (i.e. the computation) as ‘Interface Uniformity’, which holds that ‘the syntax-semantics interface is maximally simple, in that meaning maps transparently into syntactic structure; and it is maximally uniform, so that the same meaning always maps onto the same syntactic structure’. This bias inherent in much current theorizing provides a standardized way of encoding the data, but does not appear to have any strong empirical motivation; nor is the encoding it provides a particularly elegant or transparent one. Conceptually it corresponds to a picture where syntax ‘includes’ interpretation, in the sense that all relevant semantic information finds itself translated into syntactic structure. In contrast, we agree with Culicover and Jackendoff (2006: 416) on the idea that interpretation is ‘the product of an autonomous combinatorial capacity independent of and richer than syntax’, ‘largely coextensive with thought’, which syntax simply restricts in crucial ways.
Linguistic meanings are merely an input to general inferential processes; the linguistic categorization of the conceptual space encoded by lexical items does not correspond to ‘meaning’ itself but rather to a restriction of the inferential processes producing it. Sperber and Wilson (1986: 174) provide a particularly compelling discussion of the point that linguistic expressions only denote because of their inferential associations: ‘Linguistically encoded semantic representations are abstract mental structures which must be inferentially enriched’. In such a model, the well-known indeterminacy of linguistic meanings becomes a key property of successful communication:
A linguistic device does not have as its direct proper function to make its encoded meaning part of the meaning of the utterances in which it occurs. It has, rather, as its direct proper function to indicate a component of the speaker’s meaning that is best evoked by activating the encoded meaning of the linguistic device. It performs this direct function through each token of the device performing the derived proper function of indicating a contextually relevant meaning. (Origgi and Sperber 2000: 160)
Note that we disagree with Culicover and Jackendoff (2005) on the model of syntax to be adopted. Our analysis depends on a representational version of minimalism, roughly in the sense of Brody (2003). Crucially, the LF primitives we employ are independently available within a minimalist grammar as defined by Chomsky (1995), and in this sense the approach we take is compatible with Chomsky’s model. In fact, we would argue that our views on lexical variation and on interpretation are the simplest construal of Chomsky’s (2000b) proposals, as summarized above – much simpler than other current approaches, and in this sense closer to the core of minimalism and of the biolinguistic programme.
Therefore, any theory maintaining a functional/lexical divide must define the boundary between the two – which is a far from trivial task. The domain of spatial relations and of events involving them is a case in point. Spatial relations are covered by prepositions (or particles in their intransitive use), among other items. In particular, prepositions/particles can combine with elementary verbs to lexicalize events with a spatial component; for instance, English has put down (the book), Northern regional Italian has mettere giù (il libro). At the same time, Tuscan and literary Italian has a verb posare ‘put down’, and the examples could be multiplied (go in and enter in English, etc.). Particles in Germanic languages (but also in Romance, for instance in Northern Italian varieties) also allow for aspectual interpretations. If, on the basis of these, of the role they play in case systems, etc., we treat prepositions/particles as part of the functional lexicon, what should we infer about spatial primitives? Are they functional? If so, how is their relation to posare, enter, etc. (i.e. canonical lexical verbs) expressed?
As mentioned above, the answer envisaged by authors such as Kayne (2006) is that apparent variation in the substantive lexicon reduces to variation in the pronunciation of functional categories; hence the substrings lexicalized by what would traditionally be thought of as lexical categories consist in reality of a number of functional specifications – which may surface in some languages and not in others, or surface to different extents in different languages. In this way, the functional lexicon effectively spreads over considerable portions of the substantive lexicon; taking this to the extreme, one may want to say that lexical categories are but an epiphenomenon of abstract functional structure.
Since the proposal we are putting forward is that lexicons are merely ways of partitioning an abstract categorial space, we are in a way suggesting theories close to those we are taking issue with. At the same time, we consider it significant that we take the step of calling the lexical/functional divide into question, while they typically don’t. To begin with, the different approaches make different empirical predictions in the data domains they both address. Thus, we have specifically referred to Kayne (2006, 2008a, 2009) and Distributed Morphology, since we can directly compare our respective approaches with regard to such domains as fine variation in clitic structures, where we believe our model to be preferable on grounds of descriptive as well as explanatory adequacy (Manzini and Savoia 2009b, 2010).
The lexical/functional issue seems to us particularly noteworthy, because at heart it concerns the distinction between the narrow and broad language faculty (FLN and FLB). Let us assume that there is a universal inventory of concepts, and that the lexicon represents a way of realizing it. In theories in which there are in fact two inventories, one for functional categories and one for non-functional ones, it seems to us that the functional and non-functional lexicons are implicitly or explicitly apportioned to the language faculty narrowly construed and broadly construed, respectively. The reduction of the divide that we are proposing has implications not only for the more technical aspects of the theory of grammar, but also opens up the possibility that the universal conceptual repertory which is partitioned by language-particular lexicons is part of the broadly construed language faculty in its entirety. In fact, we see no reason why the grammatically relevant categories investigated here should not constitute categorizations in a domain of general cognition. In other words, what we are saying is that the existence of a functional lexicon associated with the FLN is not a matter of logical or factual necessity – and as such it should be open to scrutiny.
Given the position that we tentatively take on the matter – namely that eliminating the divide does not imply any empirical problem, and on the contrary allows for a certain simplification of the architecture of language – we may wonder why such a distinction is so prominent in linguistics. Neuropsychological literature provides much evidence, based both on recent brain imaging techniques and on more traditional language disorders and acquisition studies, that different brain areas are implicated by different conceptual clusters. The prediction is that
manipulable objects such as tools are strongly linked to motor behaviour and therefore their representational networks should comprise a significant amount of neurons in motor contexts. Animals, which are most of the time (visually) perceived rather than manipulated, should be represented by networks that partly reside in the visual cortex. (Bastiaansen et al. 2008)
Conversely, ‘assemblies representing function words remain limited to the perisylvian cortex and strongly left-lateralized in typical right-handers’ (Pulvermüller 1999: 260–1). This appears to underlie, in particular, the differential treatment of different sublexicons by aphasic patients (anomics, agrammatics, etc.). Given such results, it does not seem to us to be necessary to draw the conclusion that there is a functional lexicon associated with the computational system of natural language and distinguished on these grounds from a contentive lexicon. Another possibility is that
there is a continuum of meaning complexity between the ‘simple’ concrete content words that have clearly defined entities they can refer to … more abstract items that may or may not be used to refer to objects and actions and function words … According to the present proposal, the important criterion is the strength of the correlation between the occurrences of a given word form and a class of non-linguistic stimuli or actions. (Pulvermüller 1999: 261)
In other words, it is not so much the functional lexicon that has a special status within the architecture of the mind-brain, but rather certain concrete contents as opposed to more abstract ones.
Once freed from the burden of highly articulated inventories and hierarchies of functional categories, we can entertain a simpler syntax, much in the sense of Culicover and Jackendoff (2005). As already mentioned, on the other hand, we do not believe that levels of representations of the type proposed by Culicover and Jackendoff (2005), including rich notions such as grammatical functions, linking rules etc., are required by such a simpler syntax. Rather, the grammar implemented here is a representational version of current minimalist theories (cf. Brody 2003).
The relation of the syntax, and more precisely its LF component, to interpretation, as outlined above, is crucial in our view to understanding the role of language variation in the overall economy of the faculty of language. If our construal of syntax and its relation to interpretation is correct, the syntax restricts interpretation, but does not ‘contain’ it (Culicover and Jackendoff 2006). Thus the boundary between syntax and interpretation is a loose one, allowing for a number of different matchings of syntactic form to (inferentially determined) meaning. The looseness of this relation seems to be an essential design feature of the faculty of language, in the sense that it permits the invariant constructs of syntax to cover changing meanings. Lexical items are at the core of language variation simply because they represent the core unit of this interface between syntax and interpretation. In this sense, variation is not an accidental property of the faculty of language, and neither are the characteristics of variation that we try to outline in this study. Rather, they pretty much represent a by-product of the general design of the language faculty.
The aspect of our work which provides the title for this book (‘grammatical categories’) has to do with the redefinition of the grammatically relevant classes (i.e. the ‘categories’) of natural language. In general, we take it that the lexicons of natural languages are learnable in that lexical entries individuate natural classes. We apply this logic in particular to Romance complementizers which have the same form as wh–items (Italian che and the like) and to Romance sentential negations which have the same form as negative polarity arguments, in particular ‘nothing’ (Piedmontese nen etc.). In both cases we conclude that lexical identity of form is not a matter of homophony but reveals the sharing of deeper categorizations. This calls into question, among other things, the classical functional categories of C(OMP) (chapters 1–2) and NEG (chapters 3–4). Elsewhere in this book, we find no reason to entertain a functional category status for the so-called AUX(iliaries) have and be, which are argued just to be main verbs selecting a participial clause (chapter 6). In chapter 5 the cluster of meanings associated with Romance si and its Albanian counterpart u are reduced to a unified characterization which also holds of other morphological instantiations of middle-passive voice. Even syncretisms involving case morphology – and the functional category K(ASE) according to some (cf. Fillmore 1968; Giusti 1995), are analysed in chapters 7–8 as instances of ambiguous interpretation of the same underlying category, rather than as instances of default lexicalization. This, in turn, requires a revision of the categorizations provided by standard morphological feature systems.
It should be kept in mind that the functional structure that this book calls into question (COMP, NEG, AUX, K) is quite independent of recent cartographic proposals (see Cinque and Rizzi (2008) for an overview) which aim to provide a fine-grained picture of functional categories and the way in which they map to syntactic hierarchies. The result is an increase in the number of functional categories, yielding hierarchies of considerable complexity, which have been objected to on the grounds that they enrich the grammar by introducing a great number of new categories and orderings. Yet the same concern regarding the expressive power of the theory could be voiced for standard approaches to functional structure, since the creation of a new functional category or a new feature annotation of an existing category is not subject to any formal or substantive constraints.
In this book we propose a take on the problem which goes back to the very first models of exploded structures (Larson 1988), and even further to the very first approaches to ‘functional’ structure in generative grammar (Rosenbaum 1967 on complementation). We argue that structures are indeed atomized, in the sense that a wealth of differentiated head positions are projected under Merge. At the same time, our contention is that a considerable amount of this atomization (perhaps all) does not derive from the introduction of novel categories, but simply from the recursion of certain elementary, identical cells.
Thus, the complementizer (chapters 1–2) is not introduced as a specialized head C(OMP); rather, the clearly nominal nature of the complementizer in Romance languages (as in Germanic ones) suggests that the complementizer is the N complement of the matrix verb; in turn, this N takes the embedded sentence as its complement. This structure is as internally articulated as that of Rizzi (1997), but its internal articulation does not depend on a functional hierarchy. Rather, it depends on the recursion of ordinary nominal and sentential embeddings. Similarly, negation (chapters 3–4) is one of the earliest functional categories proposed under an articulated view of phrase structure, dating back at least to Pollock (1989). Based on evidence from Romance varieties, we propose, however, that so-called negative adverbs and heads are negative polarity elements, and, even more radically, that they participate in the argumental structure of the verb, coinciding specifically with the individuation of the internal argument position. Thus, in Romance languages, there is neither evidence for a lexicalized negative operator nor for a functional position hosting it.