10ne,   PDF
Syntax First, Words After 1

Mark Newson

0

Syntax First, Words After: a possible consequence of doing Alignment Syntax without a lexicon*

Introduction

Since its introduction in syntactic theory (Chomsky 1965), there has been much discussion about how to characterise the lexicon: does it contain only unpredictable information or are there generalisations stated in it?; is it static or dynamic?; how far can the syntactic information it contains be reduced to semantic information?, etc. Some of these points of view lead to an extended view of the lexicon in which certain syntactic processes are said to be carried out within it (as for example in Hale and Keyser 2002). Others have led to a reduction, or even the rejection of the lexicon, at least as it is commonly conceived of (most notably in Distributed Morphology, Halle and Marantz 1993, Marantz 1997). In the present paper I want to explore the consequences of adopting the latter position within the framework of Alignment Syntax (Newson 2004, Newson and Maunula 2006). One feature of this framework is that it operates without any notion of constituent structure: the units that the syntax manipulates compete with each other for positions defined with reference to other units in terms of linear order and adjacency1. The lack of constituent structure, in conjunction with the assumed lack of a lexicon, allows us to take the radical step of assuming that the syntax itself does not operate with any notion of word. Nothing in the syntax forms units from the elements which

• I would like to thank Szécsényi Krisztina and Bartos Huba for helpful comments on

various drafts of this paper. Their suggestions have undoubtedly improved the paper and the analyses it contains immensely and I am pleased to acknowledge their contributions. Whilst I have tried to address those comments pointing out weaknesses, it goes without saying that those which remain are entirely my own responsibility. 1 A rejection of the assumption of constituent structure whilst certainly far from standard is not at all unique to the present work. The position can be found in other frameworks such as Dick Hudson’s ‘Word Grammar’ (1984, 2010), Dependency Grammar (Debusmann, 2006) and a number of others (see Gáspár 2005, for a review). Within Alignment Syntax the position has been defended against the most obvious objections, such as ‘empirical evidence’ provided by tests for constituent structure in Newson (2004). The interested reader is directed towards this work for some discussion of the issue. The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 2

words are traditionally assumed to be bundles of, these are just arranged in a linear order. It is only after this linear order is established and is ready to be spelled out that any notion of word comes into play. Assuming something like what Distributed Morphology calls the vocabulary, i.e. a list of associations between phonological forms and the syntactically manipulated elements that they spell out, it is the act of spelling out the syntactic output in terms of vocabulary items selected for the purpose that bundles the elements into what we recognise as words. Thus a redundancy is removed from the system: there is no need for bundling to happen both in the syntax and in the vocabulary. The consequences of this main idea are far reaching into many aspects of the conceptualisation of language and language variation. This paper explores a tiny part of these in order to test for viability. The results are moderately positive enough to be encouraging, though it may be that any negative aspects reflect my own limitations rather than those of the general framework. There is obviously much more to be done. We start from a critique of some generally accepted ideas concerning the notions of syntactic category and distribution which leads us to question the standard lexical based approach. In section 3 we lay the foundations of a theory, based on Alignment Syntax principles, which proceeds without assuming a lexicon and all the notions that such a construct involves, such as syntactic categories and the pre-bundling of the basic elements which eventually constitute words. We distinguish between the syntactic processes which impose order on the basic elements and the processes which determine which vocabulary elements to use to spell out the syntactic string. In section 4 we test the proposed system on a very small fragment of English, concentrating on certain aspects of the expression of the meaning of the ‘verbal’ system. In particular we investigate the elements involved in expressing argument structure and temporally related meanings. Concerning argument structure, we concentrate mainly on different kinds of causative constructions, showing how different orderings of the basic elements affect which vocabulary items are chosen to express them. We propose that the causative light verb, make, is the result of placing the causative element at a distance from the elements which get spelled out by the root verb. The relationship between this phenomena and the expression of tense and aspect in the language is shown to be complex. Auxiliaries are claimed to be similar in many respects to light verbs, in as much as they are the spelling out of elements removed from the verbal root. Moreover, the ordering of temporal elements is shown to be interconnected to the arrangement of the elements of argument structure. Although the results are far too preliminary to be able to conclude on with any certainty, they are non-the-less interesting, if only because of the new light The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 3

they shine on old problems and the new problems that they themselves give rise to. Much work remains, but the indications are that it may be worth the attempt.

1

Category and Distribution

In the larger part of the generative tradition, syntactic categories play a major role in determining the distribution of elements in a structure. Lexical items are assigned their category in the lexicon and head positions in a structure are defined in terms of categories, allowing the insertion of only lexical items of the corresponding category. The earliest theories had these categorial positions directly determined by the syntactic rules themselves, but there has been a move away from this direct approach and under standard X-bar theory it is assumed that the syntactic rules define only category neutral positions. Nevertheless, category is still important for determining distribution in that although general positions are determined on a category neutral basis, the categories of actual position within any given structure are determined mostly by the selectional properties of the heads that the positions are related to. These selectional properties may make reference to the category of the elements which occupy the positions, a restriction called cselection. Other selectional restrictions based on semantic factors, s-selection, have also been claimed to play a role in determining the distribution of elements in a particular structure and it has even been debated whether c-selection might not reduce to s-selection (Chomsky 1986). However, this does not seem tenable in some cases, for example concerning the selectional properties of functional heads, and indeed in the Minimalist Programme the pendulum seems to have swung the other way and c-selection is deemed to set the relevant restrictive conditions for the insertion of elements related to functional and thematic heads alike. Obviously these current theories are based on much older ideas. The notion of category goes back to antiquity, though the classical grammatical approach tended to base categorisation on semantic and morphological criteria. It was the Structuralists who emphasised the relationship between category and distribution: defining categories in terms of distribution and describing distribution in terms of arrangements of the elements with given categories. It is clear however, that current approaches differ from the traditional one in that it is now recognised that distribution cannot be solely described in terms of combinations of categories. Other considerations play a role in determining the syntactic positioning of elements. Selectional restrictions are one such consideration: not every element that one might want The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 4

to categorise as a verb, for instance, can fit into exactly the same position in all structures as any other verb. For the Structuralists, this was not particularly a problem as, lacking any restrictive theory of category, differences in the distribution of elements could always be accounted for in terms of the recognition of new categories or subcategories (Harris 1954). However, there are distributional facts recognised under current assumptions that cannot be accounted for with the invention of new categories. For example the following complementary distribution patterns are contradictory if taken as determined by similarity of category: (1)

a I asked if he is here b I asked is he here c * I asked if is he here

(2)

a I asked who he met b I asked if he met Mary c * I asked who if he met

(3)

who is he meeting

The data in (1) show that complementisers and inverted auxiliaries are in complementary distribution while those in (2) show that complementisers and fronted wh-elements are also in complementary distribution. If distribution were determined solely on categorial grounds, we would therefore have to conclude that complementisers, inverted auxiliaries and fronted wh-elements all have categorial similarities. But this would predict that fronted whelements and inverted auxiliaries should also be in complementary distribution, which (3) demonstrates is not so. The standard account of this apparent paradox is to assume that while the complementary distribution between complementisers and inverted auxiliaries has a categorial explanation, in that both are assumed to occupy the same structural complementiser position, the complementary distribution between whelements and complementisers has an entirely different explanation: Chomsky and Lasnik’s (1977) multiply filled COMP filter, or whatever grammatical principle accounts for this. Under current standard assumptions then, although category has a role to play in determining the distribution of elements in a structure, there are other restrictions to be considered in addition. The majority of these can be seen as co-occurrence restrictions, such as selectional restrictions or agreement conditions. This is necessarily complex and clearly the simpler theory in which distribution is determined entirely by category cannot work. Yet the The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 5

question is begged whether there is another simpler theory that might work. Can distribution be accounted for in terms of co-occurrence restrictions without recourse to the notion of category? By and large, this question remains unexplored, though there are some approaches have attempted to do without the notion of syntactic category, most notably Distributed Morphology (Halle and Marantz 1993, Harley and Noyer 1999). If it is possible to jettison the notion of category and still account for distribution without the introduction of notions not already required, clearly this is to be preferred.

2

The Theory of Categories

A restrictive theory of categories, in fact the only one that I am aware of, coming out of early work by Chomsky (Chomsky 1965, 1970) and developed elsewhere (e.g. Jackendoff 1977, Abney 1987), assumes that syntactic categories are defined as a combination of a limited set of syntactic features, such as [±N], [±V], etc. The restrictiveness of this approach not only concerns the possible number of categories that it predicts, but also the predicted relationships that hold between the defined categories. Thus adding another binary feature [±ƒ] not only doubles the number of possible categories, but it also states that all [+ƒ] categories will share some property in common, related to this feature, and that all [-ƒ] categories will be united (at least) by not demonstrating this property. Recent ideas in generative grammar, however, have made this theory untenable as they require the addition of a great number of categories, all of which have very little to do with each other. Heads such as Neg, Foc, Top, etc. seem to belong to categories of their own (or at least to categories with a very small number of members) which, apart from their functional nature, are unrelated to other heads2. Because of this, attempts to work within a restrictive theory of category are currently few and far between. Newson et al. (2006) presents such an attempt in which various functional heads are analysed as belonging to a limited number of categories, such as light verb, light inflection, etc., defined in terms of categorial features. However under these assumptions elements of 2

Bartos Huba (p.c.) points out that this is not a strong criticism if it is the case that such functional heads are not taken to be actual lexical elements themselves, but positions filled by head movement or freely inserted features. However, this is not the position taken by everyone and indeed one of the common motivations for the assumption of such head positions is to point out the existence of some language or other which lexicalises them with some independent morphological element (see Behelli and Stowell 1996 for an example of such a justification). Moreover, as Bartos himself admits, not all functional heads can be treated as non-lexical, negation being an obvious example. Even if these are relatively few in number, they pose problems for a restrictive theory of category.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 6

similar categories are hierarchically ordered in ways which reflects their linear ordering, though nothing in the theory predicts this and indeed, as most functional elements share the same category, it would be impossible to account for the distributional facts in terms of syntactic category. Ultimately, then, the notion of syntactic category plays very little role in the syntax of these elements indicating that we might as well abandon it as far as they are concerned. But if such elements do not need syntactic categories, it becomes an even more attractive idea that we drop the notion entirely and have distribution of all elements determined on non-categorial grounds. In the next section, we will lay the foundations of such a theory.

3

A Categoryless Theory of Distribution

There are a number of current theories which have also abandoned the notion of syntactic category. For example, Distributed Morphology (Halle and Marantz 1993) has argued against the standard idea of a lexicon, distributing the morphological processes commonly thought to operate in it to various other places in the syntactic system. Consequently, the idiosyncratic information said to be stored in the lexicon also has to be distributed elsewhere or abandoned altogether. For example, phonological forms traditionally associated with lexical elements are stored in a language specific vocabulary which links forms with bundles of features, the idea being that the syntactic system manipulates the features and the forms are inserted late in the process, after the grammatical feature arrangements have been determined. Idiosyncratic semantic information, on the other hand, is stored in the encyclopaedia, part of the semantic component. The categorial aspects of the lexicon, being neither phonological nor semantic, are not stored in the vocabulary or the encyclopaedia, and as there is nowhere else to put this information, the theory runs on the assumption that it is not necessary. The approach I will outline here is similar in a number of ways, but differs in two important respects. First, the approach is linear rather than structural, based on the framework of Alignment Syntax in which elements compete for ideal positions defined in terms of linear order and adjacency to each other. The second difference is that whereas Distributed Morphology inserts vocabulary items into designated ‘word’ positions defined structurally, because of the lack of structure in an Alignment system, there are no predefined word positions prior to vocabulary insertion: it is the insertion of vocabulary items themselves which gives rise to the appearance that there are ‘word positions’.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 7

3.1 Background Assumptions Alignment Syntax is an optimality theoretic approach to syntax in which constraints are limited to faithfulness, specifically parse, and alignment families. The alignment constraints align targets to hosts, taken from the input, in terms of ordering and adjacency. Thus for any target t and host h there can be three alignment conditions: (4)

tPh tFh tAh

target precedes host target follows host target is adjacent to host

The ranking of the constraints for any target and host will determine the optimal positioning of the two relative to each other. Obviously, if tPh out ranks tFh, the target will optimally precede the host, though not necessarily adjacently. Adjacency is satisfied regardless of order and thus if this is highly ranked, the target will prefer to be as close to the host as possible, either preceding it or following it depending on the alignment conditions of other elements. If the dominant ordering constraint out ranks the adjacency constraint, competing elements will remain on their preferred side, but losers will be positioned at a greater distance from it. This gives rise to two basic patterns of distribution: elbowing, in which a target prefers to remain on one side of the host no matter how many other elements intercede between them and side swapping in which a target will appear on whichever side of the host it can get nearest to it. Elbowing follows from a high ranking of the ordering constraints (tPh and tFh) and side swapping follows from the high ranking of the adjacency constraint. An example of elbowing is the case of non-subject dependents on verbs in English, where dependents all want to follow the verb but some are elbowed to further positions by others: (5)

a the letter was posted on Monday b I posted the letter on Monday (*on Monday the letter) c I posted the company the letter on Monday (*the letter the company)

Prepositional modifiers follow the verbs they modify, but are elbowed by direct objects, which in turn are elbowed by indirect objects. All dependents however stay following the verb indicating that it is more important to maintain order than adjacency.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 8

An example of side swapping can be seen in the behaviour of the Hungarian preverb which, as its name suggests, prefers a position in front of the verb. However it is in competition for this position with a number of elements including the negative marker and foci. When any of these appear, the preverb swaps sides and appears following the verb, preferably adjacent to it: (6)

a

elment János prev-go-past János “János left” b nem ment el János (*nem elment/el nem ment) not go-past prev János “János didn’t leave” c JÁNOS ment el (*JÁNOS elment/el JÁNOS ment) “it was János who left”

These data suggest that it is more important for the preverb to obey the adjacency condition than the dominant ordering one. Targets and hosts may be single input elements or sets of input elements, termed domains. Domains are defined over sets of input elements which share a given property. For example, all the input elements related to a root predicate may be a domain to which member and non-member elements are aligned. Ordering with respect to a domain means preceding and following every element of the domain (i.e. being first or last in the domain – see Gáspár (2005) for the origins of domains in Alignment Syntax). Adjacency to a domain generally requires a target to avoid being surrounded by the domain and hence it will appear at the edges of the domain (Newson and Maunula 2006). The notion of domains was introduced to capture fronting phenomena which was difficult to account for in terms of individual alignments. While it is in principle possible to describe the fronting of an element through the assumption that its adjacency requirement to a given host is lower ranked than other elements competing for the position in front of the host, and hence it is elbowed to the front of all its competitors, certain phenomena resist this analysis and suggest that some elements have a strong requirement to be in front of a set of elements. A fronted wh-element, for example, does not conform to a weak requirement that it be adjacent to a predicate, but strongly requires to front its scope domain, i.e. the set of all elements related to the interrogative head. Yet there are other phenomena which, although they do not seem to be the result of alignment to a particular host element, cannot be described as a The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 9

requirement that they front a set of elements either, for example, second position phenomena. Some second position phenomena may be the result of a requirement that the second position element immediately follow the element in first position. But this is only feasible if the first element is uniformly identified. This is not often the case. German V2 phenomena, for example, has the verb following the subject, the topic or the wh-element, whichever happens to be required to be first. It would miss the generalisation if we required the verb to follow the wh-element or the topic or the subject, in that order of preference, given that this order is already stated for the fronted elements in the ranking of separate constraints (i.e. the requirement that the wh-element be first outranks the requirement that the topic be first, which in turn outranks the requirement that the subject be first). It would be preferable to arrange things so that the verb is required to be second no matter what is required to be first. We can envisage such second position phenomena as the result of two requirements: that the element precede the relevant domain, but that it not be the first element of that domain. This requires the introduction of an ‘antialignment’ constraint which is violated if an element is placed in a particular position relative to another element or domain. Consider the following constraints: (7) xPD violated by every member of D which precedes x x*PD violated if x precedes every member of D Note that the anti-alignment constraint x*PD is not simply the opposite of xPD, requiring x to follow every member of D. This condition is already captured by the opposite ordering constraint xFD (x follows every member of domain D). Thus anti alignment constraints do not always redundantly impose the same requirements as other existent constraints3. Consider now the ranking

3

Ordering constraints concerning individual elements may be different. x*Py seems to impose the same requirement as does xFy as both are violated if x precedes y. However, there is no other constraint with the same effect as an anit-adjacency constraint, x*Ay, requiring that x not be adjacent to y, and so it makes sense to extend the notion of antialignment constraints to this case. We therefore have the following classification of alignments:

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 10

of these two constraints with respect to each other. Suppose the domain consists of three members a, b and c, which are ordered abc by independent constraints: (8) xPD x*PD
xabc

axbc

abxc abcx * With the precedence constraint out ranking the anti precedence constraint, it is clearly better for the element x to precede the whole domain D and hence we have a fronting effect in this case. Now consider what happens if the rank order is reversed: (9)

x*PD xPD xabc

axbc

abxc abcx *

The previous winning candidate is now ruled out and so x does not come first. However the lower ranked precedence constraint still has relevance and it prefers x to be as near to the front of the domain as possible. Thus, ranking the anti-precedence constraint above the precedence constraint gives us the required second position phenomenon we are seeking4.

alignments xPy: x precedes y xFy: x follows y xAy: x is adjacent to y xPD: x precedes domain D xFD: x follows domain D xAD: x is adjacent to domain D 4

anti-alignments (redundant) (redundant) x*Ay: x is not adjacent to y x*PD: x does not precede domain D x*FD: x does not follow domain D x*AD: x is not adjacent to domain D

The relevant ranking of the subsequence constraint and the anti-subsequence constraint will give rise to ‘second to last’ phenomena. We will see later that such phenomena may exist, though the details are more obscure.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 11

3.2 Input In general Optimality Theory it is usually assumed that linguistic variation is accounted for solely in terms of constraint ranking and that the same set of candidate structures compete for optimality in every language (the richness of the base hypothesis, Prince and Smolensky 1993). This is not easy to reconcile with standard views of the lexicon and the idea that input elements are selected from the lexicon. If the lexicon contains idiosyncratic information and such idiosyncratic elements form the input for each competition, competitions will vary to the extent that lexicons vary. The assumption that there is no lexicon provides another view of the input which enables a more straightforward take on the richness of the base hypothesis. The input rather than being made up of lexical items is instead constructed from more abstract elements: the ‘conceptual units’ (CUs) which are combined by the grammatical system and then spelled out by vocabulary items. At this more abstract level, all languages can be viewed as equivalent as it seems reasonable to assume that the set of abstract CUs is universal5. Languages will differ in terms of the syntactic organisation of CUs, though of course they will also differ with respect to their vocabulary items. This difference, coming at the end of the syntactic process, is not so invasive as a supposed lexical difference which imposes conditions on the input itself and therefore on the subsequent syntactic processes and eventual selection of the optimal syntactic arrangement. Indeed, from this perspective the syntax has a certain amount of influence on the existent vocabulary, as CUs which are never placed in proximation in a language will not be spelled out as a single vocabulary item and hence there will be no such item in that language. Given another syntactic arrangement where the CUs are more proximate, there may well exist a vocabulary item to spell out this combination. For example, if a possessive feature is placed in the vicinity of a tense element, this may be spelled out as a single ‘verb’ and the language will have a possessive verb. If the possessive feature is not placed local to the tense, the two cannot be spelled out together as a single item and the language will fail to have a possessive verb, but will express possession somehow else. For example, in 5

Bartos Huba asks about colour term variation in languages. However the absence of a one to one relationship between vocabulary items and CUs is no indication that the CUs are not universal. I seriously doubt the notion that a speaker of a language which has fewer basic colour terms lacks the conceptual elements of a speaker of a language with more of them. If this were true, I, as a native speaker of English would presumably have difficulty in comprehending the difference between Hungarian piros and vörös (corresponding to English red). I do not. Obviously languages differ in terms of the set of vocabulary items they have for spelling out CUs. This variation does not however demonstrate variation at the conceptual level.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 12

Hungarian the possessive construction involves a dummy verb expressing the tense and a possessive marker on the possessed item: (10)

van egy kocsim is a car-1s. “I have a car”

In English the arrangement seems to be |tense|+|possessive|+|possessed| and the possessive verb spells out the first two features. In Hungarian the arrangement is |tense|+|possessed|+|possessive| and the verb is the spell out of the first feature only, the possessive being expressed as part of the possessed. If we envisage the input as taken from the universal set of CUs a question is raised concerning what the syntactic constraints actually refer to. Obviously there would be far too many constraints if they are in a one-to-one correspondence with individual CUs. But with no lexicon and subsequently no syntactic categories, the question is: what of a more general nature can constraints refer to? Following Distributed Morphology, we can recognise two basic types of CU: those which represent descriptive semantic content, or roots, and those which carry more functional content (Harley and Noyer 1999). The first group is a large but syntactically fairly homogenous set. I will assume that this set is extendable via combination of a basic set of CUs and, indeed, that most root CUs are made up of such combinations. Thus roots may be more or less complex depending on how many basic CUs are involved. In this paper I will not be concerned with these combinations and will treat roots as single elements. However, the idea that some roots are simpler than others will play a role in the process of vocabulary insertion, especially in cases where a ‘dummy’ is required. Depending on where they are positioned in an expression, root CUs make up what are traditionally called nouns, verbs and adjectives. The amount of overlap between these categories shows that the same root can be positioned in a number of places (how it gets realised in terms of a vocabulary item is obviously influenced by its relative position to other elements). We can therefore envisage a range of constraints that align roots to certain functional elements to define their basic positions. A root aligned to a determiner will be realised as a ‘noun’ and one aligned to tense will be realised as a ‘verb’, though these category labels play no role in the syntactic system and are purely epiphenomenal6. 6

There may be some roots that do not combine with certain functional elements, for example it is hard to think of what the combination of tense and the root disc would

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 13

The second set of CUs is smaller in number and much more syntactically heterogeneous. These relate to what in Distributed Morphology are termed fmorphemes (Harley and Noyer 1998) or functional elements in other frameworks. As Borer (1984) pointed out, it is these elements which syntactically define linguistic systems in that they are the locus of syntactic differences between languages. Moreover, in terms of distribution, functional CUs appear to be more restricted than roots, often having very fixed positions relative to the roots whose nature they determine. It is reasonable therefore to assume that there are constraints governing the distribution of each of these CUs, or in some cases for small groups of them, those relating to tense or number for example7. This would account for why these elements are difficult to categorise syntactically as in reality they are treated individually by the syntactic system. In general we can envisage an input made up of CUs, both root and functional, plus a specification of the semantic relationships between them: which roots are associated with which functional CUs, etc. These features are then manipulated by the general syntactic processor, which merely imposes orderings on them (and, if we consider deletion an option, on all possible subsets of input CUs), and the resulting set of candidate expressions is evaluated by the alignment and faithfulness constraints. The winning candidate is the optimal ordering of the optimal subset of CUs taken from the input.

7

produce. But this is a matter for semantic interpretation not for the syntactic system: in so far as it is conceivable to disc something, perhaps meaning to turn it into a disc, the combination of this root with a tense is possible and there may be a language which spells it out as a verb. Other languages with no such vocabulary item will spell the combination out in other ways, to make/turn something into a disc, for example. Cinque (1999) argues that functional elements are also ordered with respect to each other in a universally fixed hierarchy. He considers two alternative accounts for this: one that the fixed ordering has a purely structural basis with functional phrases hierarchically organised to produce the given ordering and the other that there are abstract ordering principles of a non-structural nature acting in conjunction with the structural conditions. He argues that the first alternative is to be preferred on simplicity grounds. Of course, the third alternative where the ordering is produced by non-structural principles operating without structural conditions was not considered and such a position is just as parsimonious as the one Cinque argues for. Indeed, as Bobaljik (1999) points out, Cinque’s strict structural argument gives rise to paradoxical conclusions which lead to unwarranted complexities necessitating a retreat to the less simple alternative unless a more radical position is taken on the nature of structure. It is difficult at present to argue whether a totally structure based approach or a totally non-structure based approach is simpler as complexities arising with the later are yet to be discovered. However, it is clear that Bobajlik’s paradox is avoided if we take a non-structural position.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 14

It is after this optimal ordering has been determined that vocabulary inserting takes place. We will discuss this process in more detail in section 3.4. 3.3 Argument Structure Without a lexicon, there can be no storage place for differences between predicates in terms of argument structure and hence there are no theta grids or the like. What argument interpretations are compatible with any given root, I take to be a semantic issue which plays no role in the syntax. The issue concerns the status of the oddity of expressions such as he put. My claim is that this is not a case of ungrammaticality but of semantic ill formedness (see Huang, Li and Li 2009 for a similar suggestion). Still, it has to be recognised that there are syntactic phenomena which are based on argument-predicate relationships and these cannot be entirely swept out of the system. In many languages arguments are ordered with respect to the predicate and each other. Moreover, the same predicate may be able to appear with certain arguments in one environment that it cannot in another. As Marantz (1997), drawing from Chomsky (1970), points out, some nominals do not have the same argument structure as their verbal counterparts: (11) a b c

the enemy destroyed the city the city’s destruction the enemy’s destruction of the city

(12) a b c

John grows tomatoes the tomatoes’ growth

• John’s growth of tomatoes

(13) a b c

John broke the glass

• the glass’ break
• John’s break of the glass

Matantz contends that these facts concern the contexts in which certain syntactic constructions can occur: the objects of roots such as destroy can appear in either nominal or verbal contexts, whereas those of roots such as grow can only appear in a verbal context. Roots like break demonstrate a third category in which neither subject nor object arguments can appear in the nominal context. Marantz’s characterisation of the data, however, is a little odd as it seems that it is not the object of roots like grow that is prevented from appearing in nominal contexts but the external argument. While we can have constructions The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 15

such as (14a) in which the theme appears in object position, the interpretation of (14b) shows that the allowable argument must be interpreted as the theme and not the causer: (14) a b

the growth of the tomatoes John’s growth

It appears then that it is that which licenses the causer that is banned from nominal contexts. If the only things there are to syntactically manipulate are the CUs provided in the input, then what is being manipulated in the phenomena discussed above must be CUs. I will assume that there are thematic type functional CUs which license the related arguments, by providing something for arguments to be aligned to and which themselves are aligned with respect to the root8. As they are not realised independently in most cases, I assume that they are normally positioned close to the root and therefore get spelled out by the root itself. There are cases where such elements get separated from the root, however, and hence are spelled out independently. This will provide an account of certain ‘light verbs’. For example, the periphrastic causative involving make can be seen in these terms: (15) a

b

…

…

John

broke

the glass

…

…

John made the glass

break

In (15a), the causative thematic CU licensing the external argument is adjacent to the root and is presumably spelled out by the root vocabulary item (this item being associated with this feature in its vocabulary entry). However, the causative CU in (15b) is separated from the root by the theme and hence it cannot be spelled out along with the root. For this reason, then, it is spelled out 8

Given the difficulties linguists have faced in discovering a universally accepted list of thematic roles, it is unlikely that the thematic CUs I make use of here are correctly labelled. It may be that a more accurate system can be devised making use of more event based notions, such as proposed by Ramchand (2008). I put the issue aside for now, pending further investigation.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 16

separately and is realised by the vocabulary item make. We will discuss the choice of vocabulary item in the next section. 3.4 Vocabulary Insertion If the output of the syntactic system is a linear string of input CUs lacking any structure, neither syntactic nor morphological, it is obviously a crucial question how vocabulary items are chosen to spell this string out. There are a number of principles involved in this and we will spend some time discussing these here. In most approaches which assume a late vocabulary spell out of an underlying abstract syntactic representation, an important principle is that vocabulary items are chosen on a ‘best fit’ basis. What this means is that if there are syntactic features xyz to be spelled out, a vocabulary item associated with the features xyz will be selected, but in the case that there is no such item another will be chosen which is considered to be the closest available. There are a number of possibilities, however, for determining what counts as the closest available vocabulary item in cases where an exact match is not possible. In Distributed Morphology a vocabulary item can be selected only if the specified features of the vocabulary item is a subset of the bundle to be spelled out and hence the one that is considered the best fit is the one that is associated with the largest subset of features needing to be spelled out. Thus vocabulary items tend to be underspecified for features. However another view is taken by Caha (2007) working within the Nanosyntax framework. Arguing from empirical observations concerning Czech Case paradigms, Caha claims that it is better to assume that vocabulary items can be selected only if they are specified for a superset of features of those to be spelled out, the notion of best fit then being equated with the vocabulary item associated with the smallest superset of features needing to be spelled out. In order to achieve the same results in Distributed Morphology, working with underspecified vocabulary items, additional mechanisms such as Fusion and Fission are required. In Nanosyntax such complexities are not required as the assumption is that vocabulary items can be inserted under non-terminal nodes of a structure and thus cover a more dispersed range of specified terminal features. In Alignment Syntax, given that there is no organisation of output expressions into structures, it is a natural consequence that vocabulary items will be inserted to cover a range of CUs and there is no need for any extra processes to collect these into one position. Indeed, as there is no constituent structure there can be no such positions properly defined. It is the very process of vocabulary insertion that groups the features into bundles rather than there The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 17

being syntactic processes which do this, redundantly, prior to vocabulary insertion. In this sense, the system is more like Nanosyntax than Distributed Morphology, though of course the former is also committed to constituent structure defining possible insertion points for vocabulary items, even if these positions are not necessarily terminal. Therefore, it seems that the conditions are more suitable in the present framework for operating with over-specified vocabulary items. Indeed, other assumptions within Alignment Syntax also argue in favour of this position. It is assumed that faithfulness violations are restricted to the ‘Parse’ variety, in which input elements are not visible in the output. Thus there is no syntactic ‘insertion’ process, including elements in the output which were not present in the input. Yet there are many syntactic phenomena which give the appearance of insertion, the use of pleonastic elements being the most obvious. We can handle pleonastics without assuming Fill violations if we account for the phenomenon at the point of vocabulary insertion9. The idea is based on Grimshaw’s (1997) analysis of do-insertion. In this, the inserted word do is not considered to be a dummy auxiliary: a lexically specified semantically empty verb. Instead the fully meaningful verb is selected for insertion, and its semantic content is simply ignored. This accounts for the fact that pleonastic elements are always identical to meaningful words in a language. If there really were semantically empty lexical items one would have to wonder why they do not have completely different forms to other lexical items. Adopting these ideas, we can achieve the effect of insertion through the selection of vocabulary items which are specified for a greater number of features than they are used to spell out. Of course, the principle of ‘best fit’ governing the selection of vocabulary items will always mitigate in favour of ‘least insertion’, i.e. the selection of the vocabulary item which has the fewest number of extraneous specified features. Thus, this will limit ‘insertion’ to those cases where there exists no vocabulary item to spell out the exact set of CUs required to be. For example, if certain functional CUs need to be spelled out independently of a root as they are not contiguous with it, but the only vocabulary items associated with those CUs are also associated with roots, then one of the root vocabulary items must be selected and its root content be treated as over-specified. This provides some account of which vocabulary items tend to be used for pleonastic purposes, along similar lines to 9

In OT, faithfulness constraints sanction differences between inputs and outputs. These differences can come in two types: input elements are missing from the output and output elements are not present in the input. The first violates a constraint known as Parse, which is therefore like a constraint against deletion. The second violates the Fill constraint, a constraint against insertion.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 18

Grimshaw’s original account. According to Grimshaw, it is no accident that the dummy auxiliary is do as this is one of the most conceptually simple verbs – in our terms, made up of the fewest number of basic root CUs – and hence it is the best fitting item to spell out pure functional CUs such as tense. We will develop these ideas in section 4 when we discuss the treatment of other auxiliary verbs. Another reasonable principle governing vocabulary insertion to assume is that vocabulary items are selected to cover the maximum number of CUs in the output expression as possible. Thus if it is possible to spell out features xyz as a single vocabulary item it will be spelled out by this item, rather than selecting one to spell out xy and another for z. In other words, there is an economy condition on vocabulary selection: don’t use two vocabulary items when one will do! Again this argues for a superset approach to vocabulary insertion. If it were possible to ignore the features constituting the output expression, as the subset approach advocates, the most economic spell out would be to say nothing! However, if it is the features of the vocabulary items which can be ignored and not those of the output expression, economy urges the use of the fewest vocabulary items to spell out all of the syntactic features10. Clearly morphological fusion, in which a vocabulary item is associated with a number of CUs, is more economical in these terms and hence should be preferred over isolation, where vocabulary items are more in a one-to-one relationship with CUs. We might wonder therefore why there are isolating languages at all and indeed why all languages are not maximally agglutinating. The answer is that there is a play off between economy and efficiency. The more fusion there is, the more vocabulary items there must be, as each root will need to be associated with a number of phonological forms depending on which functional CUs it is fused with. More vocabulary items clearly make the process of choosing the best fit in any one case more complex and the system will be less efficient. Thus economy and efficiency pull in opposite directions and individual languages find their own equilibriums in this tug of war. Some tend towards economy at the expense of efficiency while others opt for efficiency over economy. There seems to be no single most satisfactory solution and hence languages vary on they one they settle on. I do not think 10

Bartos Huba raises the question of whether the vocabulary insertion process may not be seen in OT terms, given that it involves a competition between candidate vocabulary items and a process determining the best one. This is possible and indeed I have pursued the idea for the choice of vocabulary insertion in the context of English modals (Newson 2008). The assumptions I made in that paper however differ slightly from the ones I am making here and it remains to be seen whether the two sets of assumptions can be satisfactorily unified.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 19

that there is any account for why one language opts for one solution, any more than there is an account of why different languages make different choices about the associations made in their vocabularies11. A final thing to consider concerning vocabulary insertion is that if the output expression is simply a linearly ordered set of CUs prior to vocabulary insertion, where does vocabulary insertion start? How do we choose which groups of features to spell out with which vocabulary items? One fairly obvious principle in operation here, without which there would be little point in linear ordering and hence no possible account of syntactic phenomena, is that only contiguous CUs can be spelled out as a single vocabulary item. If it were possible to spell out non-contiguous CUs as a single vocabulary item, they might just as well be placed in any order prior to vocabulary insertion. We would then need extra mechanisms to place the vocabulary items in the grammatical order and hence we would undermine the very idea of late lexical insertion. Even with this assumption, though, it still remains to be made more precise how we select which CUs to be spelled out by which vocabulary item. Supposing we have a string of features abcd. Do we start at the front and look to find vocabulary items to spell out a or ab or abc? Or do we start at the end to search for vocabulary items to spell out d or cd or bcd? Obviously different strategies will produce different results. If one looks at the issue from a purely phonetic point of view, it might be natural to assume that vocabulary insertion should follow a left to right pattern, reflecting the order of the phonetic exponents12. However, syntax is as connected to semantics as it is to phonetics, traditionally being conceived of as a bridge between the two. From a semantic point of view it perhaps makes more sense for vocabulary insertion to be root based, highlighting the semantic centrality of roots. Such a strategy would start with the roots and look to find the largest number of contiguous features around them to be spelled out by a vocabulary item. Any remaining functional CUs will then have to be spelled out separately by other vocabulary items. The evidence presented below appears to support the assumption of the root based strategy, the reason for which is, I think, not difficult to comprehend. Syntactic alignments affect those elements which are related in the input and input relationships reflect semantic relationships. Thus the tendency is to cluster semantically related elements together in the linear output. A purely linear strategy of vocabulary items would add a degree of chance to whether semantically and syntactically related elements get spelled out on different and otherwise unrelated vocabulary items. We will see that 11 12

Thanks to Szécsényi Krisztina for pointing out the need for this discussion. Thanks to Bartos Huba for bringing this point of view to my attention.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 20

semantically and syntactically related elements do indeed get spelled out on different vocabulary items on occasions, but it is rare that an element that is semantically related to one root, and therefore positioned in terms of an alignment to that root, gets spelled out with a separate root which just happens to be near by. Far more common is the situation in which related elements get spelled out by a single lexical item or if that is not possible, one of them is spelled out by an independent vocabulary item of its own, reflecting the linear placement of that element with respect to its related elements. Semantic encoding is therefore more straightforwardly represented, an obvious advantage. Now that we have set all of the parts of the theory in place, it is time to demonstrate how it works on a few example analyses. My intention is not to be comprehensive here, dealing with a broad range of issues and showing how ideas already captured in other frameworks can be rendered in this one. Instead I will content myself with a more random approach, following leads which appear interesting or which challenge the system. The approach is therefore exploratory rather than expository.

4

Some Example Analyses

I will start this section by looking at the organisation of the thematic CUs, using English as a test case. After this we will turn to an account of tense and aspect and its interaction with the thematic CUs. 4.1 The event domain Most of the time, thematic CUs are spelled out, along with others, by the root vocabulary item. Thus their positioning cannot be directly observed. However, under the assumption that the arguments they licence are aligned with respect to them, the positioning of the arguments gives indirect information about how the CUs are ordered13. One of the basic observations concerning the arrangement of arguments in English is that, in normal circumstances, the root precedes all but one of them. The ‘external argument’ precedes the root. But, although it is possible to predict which argument will be external, given a set of them, the mapping between arguments and grammatical function is not a simple one-to-one type. 13

The question of learnability is raised at this point, given that what must be acquired is only indirectly represented in the data available to the learner. To address this issue properly however, more work would need to be done concerning the amount of language variation in this area. It may be that there are certain universal aspects to the system that can therefore be assumed not to be acquired. Any variation, however, must be assumed to open options up in the system and these are loci for learning.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 21

For example, the theme is a typical post root argument, but when it is the only argument it will appear in the pre root position. There are two aspects to this distribution pattern. One concerns the arguments’ order with respect to each other, something which is often discussed in terms of a thematic hierarchy, but from our point of view will be represented as a linear ordering imposed on the thematic CUs. The other concerns the root’s position with respect to the set of arguments. Abstracting away from the details of the arguments for the time being, we can represent the position of the root in the following way: (16) a b c

arg √ arg √ arg arg √ arg arg

Clearly in the domain made up of thematic CUs related to a particular root, let us call it the event domain (De), the root is in second position. We have seen how second position phenomena is achieved by ranking the anti-precedence constraint above the precedence constraint (which itself dominates the subsequence constraint, so that the main ordering principle is precedence, but not of the whole domain). Thus we can propose the following constraints: (17) √*PDe: violated if the root precedes every member of the event domain √PDe: violated by every member of the event domain that precedes the root The following simple tables show how the root is positioned with respect to its arguments in the cases in (16): (18) √ arg
arg √

√*PDe √PDe

• !

√*PDe √PDe √ arg arg

• !

arg √ arg

arg arg √ !

(19)

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 22

(20) √ arg arg arg
arg √ arg arg arg arg √ arg arg arg arg √

√*PDe √PDe

• !
• • !
    • !

The tables speak for themselves and do not need further comment. The ordering of the thematic CUs themselves can be handled by a ranking of precedence constraints. I will assume that the basic order of arguments is essentially that proposed by Grimshaw (1990) (agent > goal > theme) and rank the following constraints accordingly: (21) |agent|PDe: violated by every member of the event domain which precedes the agent

precedes the goal

precedes the theme As the thematic CUs generally group round the root, I will also assume a set of root adjacency constraints which will not be discussed here, but will be relevant to the analysis of the interaction of the tense and aspect CUs with respect to the event domain. The table demonstrating this system in full is rather long, and becomes even longer the more input CUs are involved. But if we concentrate on those possible orderings in which the root is in second position, which we have already demonstrated, we can cut this down substantially: (22) √*PDe √PDe |agent|PDe |goal|PDe |theme|PDe
|ag| √ |go| |thm|

|ag| √ |thm| |go|

• • !

|go| √ |ag| |thm|

• !

|go| √ |thm| |ag|

• • !

|thm| √ |ag| |go|

• !

|thm| √ |go| |ag|

• • !

In (22) all the possible orders of the arguments are given with the root’s position held stable. Any other position for the root would result in one of the two higher ranked constraints being violated to a greater extent than any of the The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 23

ones presented here and hence these could not be optimal. As all the given orders perform exactly the same on the two higher constraints, their differentiation depends on the lower ranked thematic precedence constraints. This is straightforward: it is more important for the agent to precede than it is for the goal and it is more important for the goal to precede than it is for the theme. Therefore we achieve the correct order: agent – goal – theme. Note that if the theme is the only argument it will obviously be first in the domain as there will be nothing to precede it. But it will also precede the root as the root is always in second position: (23) √*PDe √PDe |agent|PDe |goal|PDe |theme|PDe

* √ |thm|

• !

This is essentially the same situation as was demonstrated in (18). 4.2 Word order, grammatical functions and the power of the system At this point it is necessary to diverge a little to discuss some issues which arise from the proposals above. The first is that this approach attributes no proper role to the notion of grammatical function. Thus from this perspective there is little sense in talking of word order variation in terms of subject, verb and object combinations. Indeed, the approach proposes a completely different view of word order arrangements in natural languages. Essentially there are two dimensions along which languages may differ from each other in terms of word order, abstracting away from complicating issues such as the positioning of topics, foci, operators, etc. Along one dimension there are at least four possibilities: root first, root second, root last and root second from last. These are produced by the following rankings of the alignments: (24)

rankings precedence > subsequence and precedence > anti-precedence precedence > subsequence and anti-precedence > precedenct subsequence > precedence and subsequence > anti-subsequence subsequence > precedence and anti-subsequence > subsequence

order √ arg arg … arg √ arg … … arg arg √ … arg √ arg

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 24

If the notion of subject has any reflection in this system it is in those languages in which one argument differs from the others in that it is on the other side of the root, i.e. in root second and root second from last languages. In root first and root last languages there is no positional way of distinguishing between subjects and other arguments. Of course, arguments may differ in terms of their thematic interpretation or their Case forms. But these must be independent considerations and not determined by grammatical function. The same is in fact true of root second and second from last languages and hence the appearance of a ‘subject’ in these languages must be taken to be epiphenomenal. Despite the epiphenomenal nature of grammatical functions attributed by this approach, it is clear that the predicted word orders correspond to attested languages traditionally described in terms of the order of grammatical functions: (25) language type √ arg arg … arg √ arg … … arg arg √ … arg √ arg

traditional description VSO or VOS SVO SOV or OSV OVS

The second dimension of word order concerns the ordering of the arguments in terms of the thematic alignments. Here things are more complex as it is not entirely clear whether Grimshaw’s thematic hierarchy should be taken to be universal. If it is universal, then the OT account given here does not predict this without stipulation that the constraints involved have a fixed ranking. However, although the data are far from clear and there is much more work required to determine the facts let alone their explanation, there are a number of considerations which would argue for assuming that the thematic hierarchy is not universally fixed and hence that the constraints given in (21) are subject to re-ranking. The first of these concerns the nature of root first and last languages. As discussed, there is no proper way in terms of position to distinguish between grammatical functions in such languages. Yet traditionally a distinction is made between VSO and VOS and between SOV and OSV orders. What is the basis of these distinctions? The issue is complex because details vary from language to language. In Malagasy, for example, a well know case of a VOS language, what is taken as the subject seems to be more topic-like (Pearson 2007), whereas other VOS languages define the subject in terms of its Case form (nominative/absolutive). In as much as there are VOS languages which The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 25

define the subject in thematic terms, similarly to English, then the order of thematic elements is not universal. In some languages the agent precedes the theme and in others it follows. The same would be true for root final languages with differing fixed argument orders. A second, theory internal, consideration concerns languages which demonstrate no preferred argument order. In such languages, while order may be imposed on other grounds, there seems to be no requirement that an argument be ordered with respect to any other because of its thematic interpretation. Hungarian is an obvious example. Note that in these languages, it is not that the argument alignment constraints have a weak effect, demonstrating a relatively low ranking: they have no effect whatsoever. Thus unless we want to entertain the idea that constraints themselves, as well as their ranking, are subject to the learning process, there must be a way that constraints can be nullified without removing them from the system. In Alignment Syntax, this is achieved via violation of the faithfulness constraints. Targets of constraints, being input elements, may be absent from outputs, thus rendering the relevant constraints inapplicable. Obviously, however, there must be a reason why faithfulness violations are tolerated, and in standard OT terms this reason is that it enables the satisfaction of higher ranked constraints. Given the assumptions of Alignment Syntax, the only other constraints we can consider are alignment constraints, and therefore it must be the satisfaction of alignment constraints that sanctions the violation of the faithfulness constraints. The final piece of the puzzle comes from the assumption that alignment constraints come in opposite pairs: precedence and subsequence. Obviously, these impose opposite requirements on the positioning of targets and hence cannot be satisfied simultaneously. Their ranking with respect to each other will therefore determine the surface order of targets and hosts. But if both alignment constraints out rank the faithfulness constraint, the optimal solution will be to omit the target and thereby allow the vacuous satisfaction of both alignments. Given that there are languages with no thematically determined word orders, it follows that in these languages the thematic CUs are underparsed in the optimal string14. We conclude that there must exist both precedence and subsequence thematic alignment constraints, e.g. |agent|PDx AND |agent|FDx. It follows therefore that the thematic hierarchy in which the agent is first cannot be universal as there exists a constraint which favours a last position for this argument. Admittedly, this discussion only scratches the surface of relevant matters and more questions are raised than can be hope to be answered in the present 14

Note that underparsing input elements does not affect the semantic interpretation of the resulting string as this is read off the input not the output.

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 26

paper. What I hope to have addressed here however is the question of the viability of the theory in as much as it makes predictions about possible linguistic variation for which there is some support in attested cases. 4.3 Causative With a number of roots in English there are two causative constructions involving a causative thematic CU. In one of these, this CU is aligned to the root and is spelled out with it and in the other the cause CU is distanced from the root and is spelled out by the light use of make: (26) a b

he closed the door he made the door close

There are a number of interesting syntactic and semantic observations concerning the relationship between pairs such as these. The arrangement of the thematic CUs is of primary concern for us here. When the causative is spelled out with the root, the theme argument, and thus the theme CU, follows the root. But when the causative is spelled out independently of the root, the theme precedes. The preceding theme is unexpected in this case as it seems that this places the root in third, rather than in second position with respect to the event domain. A reasonable account of these observations can be given if we first consider some of the semantic differences between the examples in (26). While the meanings of these examples are obviously related, they are not identical. (26b) for example, would hardly be an accurate description of a situation in which someone walked up to a door, exerted a little pressure on it’s handle so that the door moved towards the closed position. Such a situation is what (26a) describes. The periphrastic causative, on the other hand, would be more appropriate to a situation in which someone carries out an action not involving the door at all, say for example opening the window, but that the result of this action is that the door comes to be closed. Thus it seems that there are two levels of causation here: one which directly involves the argument that undergoes the result and one which does not. Suppose these involve distinct CUs, a direct causative (|caused|) and an indirect causative (|causei|). The fact that these would be independently positioned by their own alignment conditions does not help us very much at this point as it is the root’s alignment with respect to the event domain that is the problem. However, the identification of these distinct CUs does allow us to hone the definition of the event domain more precisely. Suppose we define the event domain to be the set of thematic CUs directly related to the event The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 27

described by the root, and as such it will exclude the indirect causative. We can then define an ‘extended event domain’ to include the indirectly related thematic CUs as well. With these notions in place, we can see that the root is positioned in the second position in the event domain, but not the extended event domain. When there is an indirect causative, assuming this to be ordered in front of the other elements of the extended event domain, the root will effectively ignore this CU and appear behind the first direct thematic CU. In the following tables, the distinction between indirect and direct CUs is indicated by subscripts and the members of the event domain are in bold: √*PDe √PDe √ argd argd

• !

argd √ argd

argd argd √ !

(27)

(28) √ argi argd argi √ argd
argi argd √

√*PDe √PDe

• !
• !

(27) may represent a case of a direct causative as in (26a). The actual arguments appear surrounding this string, being aligned to their licensing CUs and hence the root can spell out the whole string as represented in (15a). (28), on the other hand, represents a case of the indirect causative, with the theme CU positioned in front of the root. The theme argument itself is positioned in front of its licensing CU, interceding between the causative CU and the root. Thus the causative CU cannot be spelled out as part of the root as the two are not contiguous. In this case, apparently the root make is the best fit vocabulary item. The selection of make as the vocabulary replacement of the causative CU is a complex issue. We can gain some understanding of the process if we consider some possible alternatives to make. First, as what needs to be spelled out in this case is not a full root meaning, consider the possibility of using very ‘light’ verbs such as be or do. These are out, however, by the fact that they are too light and their root content does not include any aspect of causation. Given that the best fitting vocabulary item is over-specified rather than underspecified, this means that minimal content roots will not be used to express causation. Therefore we are looking for roots which have causation as part of their content. Make certainly has this, assuming its meaning in ‘to The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 28

make something’ is the one that it is associated with in the vocabulary. But there are many other vocabulary items which are also associated with causation. Most of these however, such as create, kill and as we have seen close, are too over-specified. Both kill and close are causatives that are specified for a particular and fairly complex result (to become dead and to become closed, respectively) whereas make is a causative where the result is simply the existence of something. Create is similar to make in this respect, but is also associated with other meanings to do with aesthetics and the like. Of course, there are other causative roots which are used to express causation in much the same way as make is, such as have and let. But these are used to spell out specific meanings, differing in terms of the amount of force needed to bring about the caused resultant state. Presumably then, these aspects of meaning are associated with these vocabulary items to allow a specific choice in a specific context. Stating the conditions for the correct vocabulary insertion in these cases is clearly a complex task and I will not attempt to go further into the issue here. I trust that a better understanding of the conceptual make up of roots will enable us to say more in the future. 4.4 Tense and Aspect Outside of the event domain and often more remote from the root we have elements expressing verb related meanings such as tense, aspect and modality. In English there is a varied system with elements of tense and aspect either being expressed as a morpheme attached to the root, or as a vocabulary item separate from the root, traditionally termed an auxiliary. The English system of tense and aspect is very regular and thus easy to describe, though it is deceptively difficult to explain. We can get some idea of this complexity by considering the first generative account of the phenomena: ‘affix hopping’ (Chomsky 1957). In this it is assumed that the elements of aspect are inserted into a structure as an auxiliary accompanied by a suffix. Tense is inserted only as a suffix. These elements are inserted into the structure in the following order: (29) -tense (have+-en) (be+-ing) V After the elements have been inserted a transformation moves each affix backwards onto the following verb/auxiliary and thus we end up with: (30) have-tense be-en V-ing

(e.g. had been drinking)

The Even Yearbook 9 (2010), Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 2061–490X, http://seas3.elte.hu/delg/publications/even, © 2010, Mark Newson

Syntax First, Words After 29

This very neatly captures the regular surface order of the elements involved, but the difficulty is in understanding the two parts of the aspectual elements. In Chomsky’s original system, these were inserted as a single element, but then are treated as independent elements by the transformation. After the introduction of the lexicon into the grammatical system, this became particularly problematic as it is unclear whether we are supposed to take the associated auxiliary and suffix to be one or two items. If it is one item, then how is it possible that the transformation can split it and if it is two items, why do they seem to express a single meaning and how do they get inserted together? A reasonable response to these questions is to assume that one of the two elements is the central meaningful item while the other is a pleonastic element inserted for grammatical purposes. At first sight it might seem obvious that if this is so, then it is the auxiliary which is meaningful and the morpheme is merely a grammatical reflex. However, Newson et al. (2006) present arguments to suggest exactly the opposite. The grammatical motivation for the auxiliaries are similar to those that force the use of the ‘dummy’ auxiliary do. It is commonly held that this auxiliary is inserted to bear a bound morpheme which, for one reason or another, cannot attach to the main verb, for example in the presence of negation: (31) a b

-ed smile → -ed not smile → →

smile-ed

• not smile-ed

do-ed not smile

(did not smile)

Pretty much the same analysis can be given for the other auxiliaries, though in these cases the stray morpheme is the result of English being unable to attach more than one such morpheme to the root: (32) a b

-ed –en smile → -ed –ing smile →

have-ed smile-en be-ed smile-ing

(had smiled) (was smiling)

One argument in favour of this analysis is the fact that when there is no stray tense morpheme to bind, the auxiliary is not inserted, even though the aspectual morpheme is present expressing the relevant meaning: (33) a b

I saw [him smiling]

is smiling