Henk Zeevat und Manfred Stede


1 (2004)

Discourse particles are more and more investigated from a formal perspective. This is due to the fact that a proper formal understanding of them is related to a number of strategic goals. Discourse particles seem to challenge some foundational tenets of NL semantics and pragmatics as it is currently taught. A curious semantic property of particles is that they seem to stand in semantic opposition with the sentence without them, which seems to entail or implicate that the marked property is false. This goes directly against the compositional picture where the meaning of a complex is a function of the meaning of its parts (and not of their non-parts). A similar foundational question is whether particles can be regarded as presupposing whatever they express or whether they conventionally implicate that (and what that would mean). There are various systematic ways in which they seem to differ from other presupposition triggers and a proper understanding of that difference and of the way in which they do function could lead to deeper insights in the nature of presupposition and/or conventional implicature. Finally, particles are simple representatives of a class of grammaticalisation processes, and a better understanding promises progress in the understanding of grammaticalisation.
In computational linguistics, a better understanding of particles would contribute to the naturalness of automatically generated text. At the moment, intuition somehow tells us that a generated sentence is unnatural because some particle must be inserted. But it is very easy to come up with rules for inserting particles that make the text even less natural than it was without the particle. Hence, particle selection and placement have to be carefully synchronized with other generation decisions. In NL understanding, it is generally assumed that one would do better with particles as they would help in understanding what move the speaker or writer is making. This in turn has a strong bearing on inferring the discourse relation, the resolution of pronouns and presuppositions, and the meaning of the tense and aspect operators. Generation and understanding come together in spoken dialog systems, where indeed particles play a major role, since it is well known that in spoken language they are even more abundant than in writing. – All in all, progress with the particles (or their translation equivalents) seems to open up perspectives in linguistic theory and natural language processing that seem currently out of reach.
The papers in this volume, which grew out of a workshop held at the ESSLLI Summer School 2003 in Vienna, are representative of the formal interest in discourse particles, but also address other issues. The most surprising new theme is the investigation of particles by means of empirical methods like corpora and experiments. The first five papers are instances of this new development, in which Spenader and Soffner are doing classical semantics/pragmatics with new methods, and Alexandris & Fotinea and Alonso
et al. develop theoretical understanding in pursuit of computational goals. Spenader uses parallel corpora to test and refine various analyses about the Swedish particle „ju“ which is notorious for being hard to describe. The parallel corpus gives a variety of English lexical items or constructions that express different uses of „ju“. Alexandris and Fotinea give the theoretical results of applied research into politeness marking and speech acts, using both corpus investigation and experiments. While they aim for results that can be applied in fielded dialogue systems, they find that a group of particles have the double function of marking speech acts and expressing respect for the interlocutor.
Alonso, Castellon, Shih and Padro exploit the ease of recognition that particles provide for shallow processing of discourse structure. Annotation schemes are proposed and evaluated for drawing conclusions from particles and other discourse markers to the structure and intention of the utterances. Soffner develops an interesting methodology of corpus experimentation. The method consists in trying to find consequences from theoretical ideas that are observable in a corpus. The method is applied to the idea that „but“ always expresses denial of expectation. This would predict that „but“-clauses can never be marked as expected. This is tested on one such marking, by tantamount tag questions. (And then John sang, did he?) Thomas also studies „but“, and in particular reflects on the relationship between denial-of-expectation and frustrated-plan usages in different types of dialogs – where a frustrated-plan „but“ would cover cases like „John searched the entire building but didn‘t find his cat“, where arguably an expectation to be violated is at best well-hidden.
The last three papers use classical formal semantics/pragmatics methods and are targeted towards particular particles. They all represent a recent tendency in which real progress is made in the formal analysis of discourse particles. This seems due on the one hand to progress in the development of formal pragmatics and on the other to a wave of interest in these particles. Karagjosova gives the first working theory that unifies „doch“ and „DOCH“. This pair of particles seem to arise from each other (by accenting or deaccenting) and to be provided with opposite meanings: „doch“ marks the utterance as common ground between speaker and hearer, whereas „DOCH“ seems to indicate that the utterance is the negating part of the common ground. The crucial step is the adoption of the fi ne-grained distinction of Prince (1981) and Walker (1993) between evoked, inferable and brand-new. Eckhardt in a contributed squib presents a first look at particles like „dann“, „noch“, „sonst“ and „vielleicht“ (then, yet, in addition, maybe, but with particular meanings in the question environment) as they appear in questions. The question environment of particles is relatively unknown and a preliminary analysis is offered in terms of communication strategies (Büring 2003) and presupposition. Schmitz and Schröder provide a new theory of “eigentlich” (really) in which it blocks default inferences from the clause which contains “eigentlich”. The analysis is able to deal with an extensive number of examples (counterexamples seem to be entirely absent) and is formalized using Veltman’s (1996) default logic.

Büring, D.(2003) On D-trees, Beans and B-accents. In Linguistics and Philosophy 26(5),
Prince, E. (1981) Towards a taxonomy of given-new information. In P. Cole (ed) Radical
Pragmatics. New York.
Walker, M. (1993) Inofrmational Redundancy and resource bounds in dialogue. PhD thesis.
University of Pennsylvania.
Veltman, F. (1996) Defaults in Update Semantics. In: Journal of Philosophical Logic (26),