Are linguistic signs really as arbitrary as they are assumed to be, and how exceptional are semantically motivated lexical patterns? These old questions are re-addressed with the help of new quantificational data on two genetically unrelated languages, Kambera (Austronesian) and Dutch (Indo-European). The hypothesis is that semantically complex items such as expressives favor a structurally complex form, and vice versa. An examination of seven distinct types of lexical items found statistically robust semantically motivated lexical patterns. Linguistic signs appear to be less arbitrary than is commonly assumed.

This article attempts to provide a precise definition for topic and to derive most of the properties of topic from this definition. The main assumption is that the topic-comment construction is a syntactic device employed to fulfill certain discourse functions. Topic is always related to a position inside the comment. Since topic has no independent thematic role but always depends on an element inside the comment for its thematic role, it has no syntactic function of its own. This dependence relationship is subject to locality constraints.

A grammatical affix undergoing phonetic erosion is sometimes abruptly replaced by a conveniently available lexical stem with which it shares one or more phonological segments. The new affix has the phonological shape of the old independent stem, but acquires the basic grammatical function of the old affix, though it may also bring in a portion of the stem's own morphological and semantic idiosyncrasies. Because the old affixal form is eliminated, the historical process can easily be misdiagnosed as reflecting the gradual compression of an original syntactic construction which includes the relevant independent stem. Recognition of the system-renewing element in this particular type of stem-to-affix grammaticalization leads to awareness of the pivotal role of the inherited system in grammaticalizations, and to a general critique of historical models which see grammaticalization as a straightforward syntax-to-morphology compression.

This article examines stative and passive constructions in Chichewa, finding that the syntax of the two constructions is best accommodated in a model that allows argument structure changing operations, such as stative, to be distinguished from operations, such as passive, that affect the mapping from argument structure to grammatical functions. The Chichewa facts recall observations made concerning the differences between English adjectival and verbal passives, first formalized in Wasow 1977, but provide more straightforward evidence for a distinction, due to the absence of surface homophony. The paper concludes by reconsidering English adjectival and verbal passive constructions, and showing them to be morphologically distinct.

We support the theory of level ordering by demonstrating, on the basis of productive morphology and phonology, that Turkish has four lexical levels. The evidence, however, motivates modifications in the way level ordering is implemented. The first is the principle of Level Economy, according to which a form is subject to the phonology only of those levels at which it is morphologically derived. The second is level prespecification, which exempts a root entirely from early lexical levels. Level Economy accounts for systematic exceptionality, while level prespecification accounts for idiosyncratic exceptionality, to the entire phonology of given levels. These mechanisms yield analyses of facts in Turkish that prove intractable in other theories. Both rely on a structural, rather than a temporal, approach to level ordering.