Lexicalization algorithms using non-terminal spellout

In this section I will discuss the different ways in which the idea of non-terminal spellout has been implemented in the literature, and the various linguistic phenomena it has been applied to. This will serve as a background to the detailed exposition of Nanosyntax in Section 2.4.

The claim that lexical insertion can only target terminals is the most widely shared assumption of syntactic theories that posit a hierarchical tree structure. This claim, however, has not remained uncontested. Overt and covert pronouns, for instance, are often thought to correspond to an entire phrase even in analyses that otherwise use only terminal spellout. Jackendoff (1977), for example, argues that the English pro-formsdo so andone correspond to V^′ and N^′ respectively, and they are incompatible with V and N complements because those are internal to V^′ and N^′. Uriagereka (1995) and Corver and Delfitto (1999) argue that (clitic) pronouns are D-elements that take a pro-NP complement.

The idea that a lexical entry can spell out an entire constituent is taken up in Weerman and Evers-Vermeul (2002) and Neeleman and Szendr˝oi (2007) to explain the distribution of pronouns.

Weerman and Evers-Vermeul (2002) examine Dutch pronouns and establish two groups: those that can and those that cannot occur with other DP-internal material such as numerals, adjectives an nouns. They argue that with both types of pronouns, there is an N present in the syntax, and the difference between the two pronoun types lies in the amount of structure spelled out by pronoun.

Pronouns that can co-occur with other DP-internal constituents sit in D, leaving N, Num and spec, Adj free to be spelled out by other lexical items. Pronouns that cannot co-occur with other

10This is reminiscent of the way Butt posits a unified lexical item for light verbs and their corresponding full verbs.

Note, however, that while Butt’s generalization has cross-linguistic validity, not all noun classifiers of all languages have a noun use.

11Wilkins proposes an analysis of the relevant noun classifier constructions in the framework of construction grammar. Given the way construction grammar works, for him the two positions do not correspond to two projections in the functional sequence. However, in the generative framework pursued here, this is the only sensible interpretation of his proposal.

DP-internal material, on the other hand, spell out the entire DP. Weerman and Evers-Vermeul further suggest that there exist pronouns that spell out KP (this gives rise to the subject vs.

object pronoun distinction) as well as pronouns that spell out some phrase between NP and DP (this is the case with pronouns that can co-occur with a D but not with other material between N and D).

Neeleman and Szendr˝oi (2007) analyze the cross-linguistic distribution of radical pro-drop.

They argue that languages with radical pro-drop i) have pronouns that correspond to KPs with a phonologically zero spell out, as in (10) and ii) these languages have no other, more specific spell-out rules for KP that could block the application of (10) by the Elsewhere Principle. (11) is an example of such a blocking rule from Dutch.

(10) [KP + pronominal, – anaphoric ]⇔ ∅

(11) [KP + pronominal, – anaphoric, 3,sg, m, acc ]⇔ hem

This account provides a principled way of explaining why radical pro-drop is attested only in languages with agglutinating pronouns. In languages with pronouns fusional for case, all pronoun spell-out rules target KP, and the application of the very general rule in (10) is blocked by the existence of more specific rules. In languages with agglutinating pronouns, on the other hand, the bare pronouns spell out some constituent smaller than KP; for instance NP (yielding pronouns agglutinating for both number and case) or DP (yielding pronouns agglutinating only for case).

An example from Japanese is given in (12).

(12) a. [N P + pronominal, – anaphoric, 1,sg ] ⇔watasi b. [K nom ]⇔ga

c. watasi-ga I-nom

Neeleman and Szendr˝oi argue that the rules in (10) and (12-a) are not in competition, as neither properly includes the other in its domain of application. (10) is more specific because it targets a larger piece of structure, and (12-a) is more specific because it makes reference to more features.

Consequently, the zero-spell out rule in (10) that produces radical pro-drop is not blocked in languages with agglutinating pronouns.

Williams (2003) uses the idea of lexical items spelling out multiple positions in an entirely different, lexicalist framework: representation theory. The detailed architecture of this model need not concern us here, so I will limit myself to a brief sketch of how Williams thinks about vocabulary items and their relation to syntactic structures. He assumes a universal set of elements arranged in a rigid hierarchical order. For the verbal domain, this is AgrS > T > Asp > AgrO > Voi > V (alternatively, T > AgrS > Asp > AgrO > Voi > V), where > stands for the complementation relation. Williams further proposes that a lexical item may realize only one element in this func-tional chain or it may realize a sequence of elements. A lexical item of the latter type is said to

‘span’ the functional chain.

As a concrete example, consider the fragment of the English spanning vocabulary for verbs given in (13) (from Williams, 2003, ch. 8., ex. 56). As the examples shows, vocabulary items may span different amounts of structure, starting and ending at different points in the functional chain.

(13) T Asp1 Asp2 Voi V might have been being killed

kills passivewas

killed has

killed been

killing modal

−s

kill

Williams’ spanning idea is also adopted by Adger (to appear). Adger applies spanning to free functional morphemes in representations which resemble Brody’s (2000a)‘telescopic’ trees. I refer the reader to the book for more details.

Lexical items spanning multiple positions in a sequence is also a central idea in Newson (2010).

Newson works in the framework of Alignment Syntax, a version of Optimality Theoretic Syntax that dispenses with syntactic structures entirely, and relies solely on alignment and parse-type faithfulness constraints to derive word order. Thus strictly speaking, it is not true that Newson works with non-terminal spell out: he has no terminals to begin with. But the spell-out mechanism he uses is best viewed as the equivalent of non-terminal spellout in a syntactic theory without structures; and he himself points out parallels with Nanosyntax on several occasions in the paper.

Newson proposes that the building blocks of syntax are conceptual units (CUs). Examples of CUs include roots, ‘agent’, ‘cause’, ‘past’, etc. He argues that there is no lexicon with pre-bundled CUs: individual, non-pre-bundled CUs serve as the input to syntax. Syntax determines an optimal order for CUs, but there is nothing in syntax that would determine which neighbouring CUs will belong to the same word. Vocabulary insertion operates on the output string of syntax.

It is Spellout itself that bundles the CUs into words, and yields the illusion that there are word positions.

Vocabulary items can spell out a span of continuous CUs. Newson argues that roots have no argument structure, rather the arguments associated with the root are licensed by thematic functional CUs. Such thematic functional CUs end up close to the root and get spelled out together with it, which makes them invisible. Verbs with irregular past tenses also spell out a span of CUs, specifically the root, the thematic CUs introducing the arguments and the past tense CU, all of which ended up in a continuous string in syntax. Some specific examples of spanning vocabulary items are provided in (14) and (15).

(14) string of CUs: past agent √ theme

| {z }

perf

spellout: had throw -n

(15) string of CUs: past cause perf

| {z } . . . theme √

| {z }

spellout: had made it break

Let us summarize this section. Although non-terminal spellout is not part of the toolbox of mainstream generative syntax, the idea has been around for a while. It cannot be emphasized enough that the concept of one lexical item spelling out more than a single terminal (or equivalent thereof) is a feature of how spellout works, and therefore it is fairly independent of the syntactic model one uses. It is, of course, incompatible with models in which syntax manipulates words with phonological information. But it is entirely compatible with any syntactic model in which syntax manipulates abstract units without phonological features and relies on late insertion. Most syntactic theories currently on the market answer to this description. Non-terminal spellout can be paired with any one of them. We have seen that this idea has been used in the mainstream generative syntax model (Weerman and Evers-Vermeul, 2002; Neeleman and Szendr˝oi, 2007), in the very much non-mainstream Representation Theory (Williams, 2003) and even in OT syntax (Newson, 2010). Non-terminal spellout is compatible with both lexicalist and non-lexicalist theories (Williams, 2003 versus the other works mentioned).

I will pair the idea of non-terminal spellout with a cartographic phrase-structure representation, a non-lexicalist syntax and an elegant syntax-semantics mapping.