Underlying free (floating) features occur crosslinguistically. These features sometime function as morphemes. Such features, like segmental morphemes, often refer to specific edges of the stem, hence they are ‘featural affixes’. They get associated with the base to be prosodically licensed. We propose to account for the association of such features through a family of alignment constraints called ‘featural alignment’ which is a featural version of McCarthy & Prince's Align (MCat, MCat). Under featural alignment, an edge is defined for a feature based on a possible licensor, which may be a root node or a mora. We argue that misalignment takes place under pressure from feature co-occurrence constraints. Thus a featural suffix may get realized elsewhere in the stem, surfacing as a featural infix or even as a featural prefix. This constraints based approach is preferred to rule-based approaches since it does not require a variety of additional assumptions needed within rule-based approaches to account for the same phenomenon. These include structure preservation, prespecification, extrationality and filters.

An analysis of phonological speech errors in Arabic, English and German is carried out with a view to probing into the organization of segments within syllables and words. Arabic slips are shown to be less structure-sensitive than English and German ones. Being absent from underlying representation, suprasegmental structures are assumed to unfold gradually in real time. The erection of hierarchical representations is claimed to be slower in Arabic than in English and German because the nonconcatenative morphology of Arabic prevents an early assignment of consonants to structural slots. In contrast, English and German words allow the early build-up of hierarchical structures because sufficient phonological information is available from the beginning of the derivation.

This paper aims to re-assess the notion DISCOURSE MARKER as it is applied to a subset of so-called apposition markers. It is argued that the classification of markers of reformulation as discourse markers alongside expressions like but and so is incoherent from a semantic point of view, since this ignores the distinction between PROCEDURAL and CONCEPTUAL meaning. Moreover, this classification is based on an account of discourse which is not only based on an insufficiently general account of context, but which is also difficult to maintain in the light of the use of these expressions in parenthetical nominal appositions. An alternative account is developed in the framework of Sperber & Wilson's Relevance Theory building on recent work on the meaning of parentheticals and sentence adverbials.

Starting from the traditional distinction between R(elational) and Q(ualifying) adjectives, we divide the former into two major subclasses: Th(ematic) and C(lassificatory) adjectives. This distinction has effects on semantic interpretation and word order. Th-adjectives obey the thematic hierarchy and C-adjectives follow patterns of subclassification of objects in different domains. R-adjectives are maximal projections mapped into specifier positions. Spanish, contrary to English, has overt raising for gender and number checking on the noun and the adjectives. R-adjectives are argued to form syntactic clusters by a series of adjunctions that satisfy antisymmetric c-command requirements and minimality constraints on movement.

This paper is concerned with the phenomenon labelled ‘metalinguistic negation’ (MN) by Horn (1985). The interaction of MN with negative polarity items, lexically incorporated negation, and accentuation, is considered in order to establish its exact scope and range of applicability. The relationships between MN and various types of implicature are advanced in support of the use of MN as a diagnostic for aspects of an utterance which are properties of the linguistic expression used. The implications of this for the status of accent are assessed. Finally, the particularly stylistic effects of MN are considered, in relation to the topics discussed.

The main aim of this paper is to discuss the claim that discourse connectives are best treated as indicators of coherence relations between hierarchically organized discourse units. It will be argued that coherence relations cannot be seen as cognitively real entities. Furthermore, there is no evidence for hierarchical organization in discourse. The intuitions underlying the notion of hierarchical discourse structure are instead explained in terms of consequences of processing a text in the search for optimal relevance. This account draws attention to a hitherto not widely discussed set of data.

Recent work in a variety of different theoretical traditions has tended to emphasize the close match between syntax and semantics (Dixon 1991; Langacker 1987, 1990, 1995; Levin & Rappaport Hovav 1991, 1992; Wierzbicka 1988). It is very easy to be left with the impression that, if only we could analyse the relevant syntactic and semantic structures correctly, this match would be total. The following are fairly typical statements:

The picture that emerges is that a verb's behavior arises from the interaction of its meaning and general principles of grammar. Thus the lexical knowledge of a speaker of a language must include knowledge of the meaning of individual verbs, the meaning components that determine the syntactic behavior of verbs, and the general principles that determine behavior from verb meaning. (Levin 1993: 11)