Most linguists who have investigated linguistic categories from a universal viewpoint have accepted the existence of two basic parts of speech, noun and verb. Other categories are found to be only inconsistently represented; thus adjective is manifested in many languages as a class of stative verb. Furthermore, individual languages often have intermediate categories such as gerund, which cannot be unambiguously assigned to a single category. We suggest here that the basic categories N and V are to be viewed as universal lexicalizations of the prototypical discourse functions of ‘discourse-manipulable participant’ and ‘reported event’, respectively. We find that the grammars of languages tend to label the categories N and V with morpho-syntactic markers which are iconically characteristic of these categories to the degree that a given instance of N or V approaches its prototypical function. In other words, the closer a form is to signaling this prime function, the more the language tends to recognize its function through morphemes typical of the category—e.g. deictic markers for N, tense markers for V. We conclude by suggesting that categoriality itself is another fundamental property of grammars which may be directly derived from discourse function.

The analysis of formulaic German binomial expressions of the form N und N reveals the following characteristics. The absence of determiners, along with other morphosyntactic and semantic anomalies, makes these phrases non-compositional and thus similar to idioms; but they differ from fixed idiomatic expressions in that their structural pattern can be productively used for the creation of new pairs. These bare binomials are not exhaustively listable lexical items, nor can they be generated by a syntactic conjunction rule. Their formal properties suggest that they are complex word-like expressions—comparable in many ways to nominal compounds and, like compounds, describable by lexical rules. Their creation and use is subject to semantic and pragmatic constraints which can be subsumed under a general principle of frame structure. They can be divided into three main types: (a) lexicalized and irreversible, (b) novel but semantically motivated, and (c) semantically unmotivated but pragmatically constrained. However, sharp lines between lexicalized and novel instances, or between semantically and pragmatically motivated types, cannot be drawn on principled grounds. Bare binomials lend support to a view of language in which the ‘generative’ and the ‘idiomatic’ components are two extremes on a continuum, rather than fundamentally distinct phenomena.

Sentence topics are linguistic constituents, with syntactic and semantic properties which single them out for a linking function in the process of relating a sentence to its discourse context. This paper explores the relation between the linguistic properties of sentence topics and their use as discourse links in sentence processing. It proposes a set of criteria for distinguishing relatively weak or strong topic NP's, based on syntactic and semantic/pragmatic properties. Syntactically defined topics include subjects and those in ‘marked’ NP positions, where the surface features define grammatical function in an ambiguous or indirect way. Implications of this proposal are drawn for several languages, and for various models of language processing.

Noun incorporation is perhaps the most nearly syntactic of all morphological processes. Examination of the phenomenon across a large number of geographically and genetically diverse languages indicates that, where syntax and morphology diverge, incorporation is a solidly morphological device that derives lexical items, not sentences. It is used for four different but related purposes; these fall into an implicational hierarchy which in turn suggests a path along which incorporation develops historically. Differences in its productivity from language to language show that this development may be arrested at any point—resulting either in the eventual disappearance of the process, or in its resurgence as a productive system of affixation.

Segmental length can be analysed in one of three possible ways: as a segmental feature, as gemination, or as a suprasegmental. These three analyses make different predictions about how length should behave in language production. Treating length as a suprasegmental predicts that it will frequently be dissociated from a segment, while the other analyses predict it will usually not be. Speech error corpora in German, Swedish, and English are examined. The data suggest a suprasegmental analysis, most probably along the lines of recent autosegmental descriptions: long segments are associated with two positions in the syllable structure. A vowel and its associated structure show different degrees of cohesiveness in different languages, so that they behave quite differently in errors in the different languages.