This paper explores diachronic change in word-formation. The phenomena of change usually attributed to the domain of word-formation range from phonological alterations of affixes, grammaticalization and the extraction of word-formation elements out of loans to the extinction of word-formation patterns due to the loss of morphological productivity. But actually, these phenomena affect different levels of word-formation, i.e. word-formation products, word-formation patterns and the word-formation system as a whole. Some of them are not even restricted to word-formation but do apply to other components of grammar as well. In this paper I will argue that the only kind of change that is specific to the domain of wordformation is the change of word-formation constraints. This change of constraints is reflected in changes of morphological productivity. Empirical evidence for this connection will be drawn from a corpus-based diachronic study of German -er-nominalization.

The paper presents a theory of the semantics und the discourse effects of German temporal dann in its sentence-initial and its sentence-internal position. The theory provides an algorithmic reconstruction of these differences on the basis of a single specification of the semantics of dann, which interacts with other semantic elements at sentence and discourse level in ways that vary with its position. The requirement which dann imposes is that the time it denotes is separated from the time which the narration had reached by a change in some contextually salient property of a topical discourse referent. What is available as contextually salient property depends on whether dann occurs at the beginning of a sentence or in the middle of it. Integration of this insight into a general account of semantic processing entails a shift of perspective in current semantic theory. Semantic processing must now be treated as incremental not only sentence by sentence but word by word.

The paper also shows how the requirements which dann imposes on context in the different positions it can occupy interact with interpretation principles involving causation, initiated action and planned action. In this connection we introduce the distinction between ,continuous' and ,discontinuous' descriptions of event sequence.

This paper provides a data structure for the transcription of handwritten texts. The data structure is modelled as an xml markup language which allows to code additional information about the graphical arrangement of a text and thus to avoid the loss of those information in the course of governing corpora of manuscripts with the computer. In particular, the presented markup language focuses on phenomena in learner texts (e.g. revisions and embedded pictures). Furthermore it allows annotating information about the writing process and that accounts for varying authors of an original text.