The functions of non-lexical feedback tokens in Swedish
Conversation is the most common usage of spoken language. In Swedish interlocutors commonly utter short feedback tokens like m, mm, mhm, a, aa, aha, ja, jaa, jaha while the other one is speaking. These tokens are usually referred to as non-lexical, indicating that they are not words per se. The conveyed meaning of these may not only be changed by altering base-morphemes and phonological operations, but also via subtle alterations of intonation patterns. This seminar presents a resource efficient method for semi-automatic annotation of these tokens as found in the Spontal corpus. It is investigated how listeners use base-morphemes and phonological operations in the course of listening to a brief story. Processing via automatic and human-driven prosodic clustering then further reveals the importance of base morphemes for the conveyed meaning as well as leaving a few tokens which cover the essential prosodic variation. These few tokens are then represented by a conventional annotation scheme which leads to acoustic correlates to understanding, agreement, interest, surprise and certainty.