Stress, pauses, pronominal types and pronominal functions in Danish spoken data

In this paper we present a study of the relation between types of third personal singular neuter pronoun and their functions in Danish spoken data where stress information is marked so that personal and demonstrative occurrences of the pronouns can be distinguished. This study confirms that there are language specific differences in the way various types of pronoun are used to refer to abstract entities in Danish compared to English data. In the second part of the paper we describe supervised machine learning experiments with the purpose of investigating the role of stress and pause information on the automatic recognition of the pronominal functions of third personal singular neuter pronouns in spoken corpora. The results of these experiments show that the inclusion of stress and pause information in the data improves classification.