In morphology, as in other branches of scientific endeavour, apparent disorder and irregularity tend to be re-analysed as underlying order. Asymmetries between form and meaning, such as allomorphy, tend accordingly to be factored into basic invariance,2 sames of meaning being interpreted, where possible, as sames of form. Alternatively, as in some of Bybee's (1985) analyses of fusional allomorphy, differences in form may be viewed as diagrammatic of differences in meaning. Allomorphic variants which cannot be made to yield to such re-analyses are commonly relegated to a kind of marginal synchronie ‘junkpile’, and are assumed to be a synchronically ‘inert’ residue of historical (usually phonological) changes. That such an approach to allomorphy can often be illuminating is not in question, but it bespeaks an essentially negative view of morphological irregularity. I wish to propose that there is room for a complementary perspective, in which the ‘irregularity’ inherent in allomorphy can be appreciated not as basically ‘inert’ deviation from a natural isomorphic relationship between meaning and form, but as an active, abstract structural property of morphological systems.

Previous functional analyses of American English inversion constructions (for example, Hartvigson & Jakobsen, 1974; Gary, 1976; Green, 1980, 1982) have recognized – either implicitly or explicitly – that inverted sentences and their canonical-word-order counterparts are semantically equivalent. None the less, in Ward & Birner (to appear), we describe a non-truth-conditional asymmetry between the interpretation of certain VP inversions and that of their canonical-word-order counterparts.2 Consider (1a) and (2b) in the following context: (1) Free elections were held yesterday in Czechoslovakia for the first time since the war. (a) The main opposition party was losing the election. (b) Losing the election was the main opposition party. (c) The main opposition party lost the election.

There is a widespread assumption within the Government–Binding theory as it has developed from the Barriers model (Chomsky 1986) that functional categories, that is, categories which play a role in establishing dependencies between parts of a sentence, as opposed to lexical categories, should be represented as heads projecting X-bar phrases. I shall refer to this as the Full Functional Projection Hypothesis (FFPH), stated informally in (1). (i) Full Functional Projection Hypothesis Any morphophonosyntactic formative which corresponds to a functional category in a given language is syntactically the head of a maximal projection.

The purpose of this article is to demonstrate that underspecification of lexical-entry forms enables us to restrict phonological theory to declarative statements about the structure of lexical items, and to avoid having recourse to feature-changing and deletion rules. The realizations of lexical items are mapped onto their underlying forms by means of filling-in, redundancy rules of two basic types, predictive and default. Predictive rules derive (at least) one feature from (at least) one other feature, given in the lexical entry form, and default rules provide a feature, if no other rule has applied. Rules are both universal and language-specific. Since all filling-in is accounted for by these redundancy rules, there is no need for a post-lexical component of the phonology.

Several restrictions on successive cyclic wh-movement appear not to be exclusively linked to general principles of the grammar, but seem to be in some sense lexically determined. It has been pointed out repeatedly that wh-movement of subjects and adjuncts out of complement CPs of factive verbs strongly contrasts with wh-movement of internal arguments out of these CPs (Rouveret, 1980; Kayne, 1981; Zubizarreta, 1982; Adams, 1985): (1) (a) *Who do you regret/understand/forget likes this book? (= Adams, 1985: (4b)) (b) *How did he deeply regret that his son had fixed the car? (c) ?Which article did you regret/understand/forget that I had selected?

English exhibits PP, AP and NP resultative secondary predicates (SPs). Italian freely exhibits PP resultatives and, less commonly, AP resultatives. This difference follows from two facts. First, resultatives, being arguments of the V except in constructions involving ‘fake’ objects (see section 4), may appear only in positions that non-predicative arguments of the V of their same category can appear in (a correlation stated in (155) below). Since English allows PP, AP and NP non-predicative arguments in the position immediately following the direct object, all three categories can also appear as resultatives in the same position. But since Italian allows only the first two types of non-predicative arguments in this position, only PP and AP resultatives can appear there. Second, Italian sentences with AP resultatives are subject to a rule of semantic interpretation by which the primary predicate must be interpreted as focusing on the endpoint of the activity it denotes (as stated in (110) below). English sentences with AP resultatives are only slightly sensitive to this interpretation rule. As a result, AP resultatives are appropriate in fewer situations in Italian. That AP resultatives are sensitive to this rule of interpretation is consistent with the fact that AP arguments of verb that appear in post-object position are marked in a number of ways.

The analysis of vowel length in Standard Scottish English (SSE) and Scots dialects has proved problematical for some time now. With most analyses the Scottish Vowel-Length Rule (SVLR) is taken to lengthen underlyingly short vowels, and the principal problem with this approach has been in giving a phonological characterization of the set of vowels which act as input to the rule. A complicating factor here is the fact that the set of input vowels seems to vary across dialects. A recent analysis by Anderson (to appear), couched in dependency-phonology (DP) representations which allow for non-specification, argues rather persuasively that SVLR is a shortening operation on underlyingly long vowels, thus resolving the problem of the phonological definition of the input set (for some dialects at least). It also goes some way towards dealing with variation across dialects by giving an explanation for the variable behaviour of the lower mid vowel [ↄ,], which participates in SVLR in some dialects but not in others.

The goal of this article is twofold. First, I explore the hypothesis that a number of regularities in the distribution of the two types of inchoative constructions with verbs of change of state in French, the superficially intransitive construction and the reflexive construction (illustrated in (1)–(3)), can be captured by an analysis whereby monovalent verbs of change of state may project the Patient argument to the subject or to the object position. When the Patient argument is projected to the subject position (as in (1a) and (3a)), the construction is unergative. When it is projected to the object position, the construction is unaccusative (as in (1b)–(3b)). Verbs of change of state in French diner as to whether they may enter an intransitive inchoative construction (1), a reflexive inchoative construction (2) or both (3).

In recent years the study of phrase structure has largely become a search for principles, parameters and other mechanisms to replace phrase-structure rules of the traditional sort. For capturing word order, one of the most common devices is the directionality parameter, especially the one which specifies a language as head-final or head-initial. It has also been proposed that languages may be specified for the direction of government (Stowell, 1981), or direction of Case- and theta-role assignment (Koopman, 1983; Travis, 1984). Often, such proposals focus only on the lowest (X′) level of phrase structure, that is to say the ordering of complements with relation to a head, but there have also been suggestions for directionality of specifiers or adjuncts (as in Ernst, 1989; Georgopoulos, 1991).

This article analyses middle constructions in English, accounting for their key syntactic and semantic properties. The analysis rests on the observation that there are certain similarities between middle, tough and recipe-context null-object constructions, such as in (1a–c). (1) (a) This bread cuts—easily. (b) This bread is easy to cut—. (c) Take bread. Cut—carefully (and arrange—nicely).