Resolving Ambiguities in Sentence Boundary Detection in Russian Spontaneous Speech

The paper analyses inter-labeller agreement within manual annotations of transcribed spontaneous speech and suggests a way to resolve ambiguities in expert labelling. It argues that the number of controversial sentence boundaries may be reduced if some of them are regarded as “zones”. We describe a technique of detecting these zones and analyse which syntactic structures are the most likely to appear in them. Though the approach is based on Russian language material, it may be applied to oral texts in other languages.