The Annotation of the HuComTech Audio Corpus in Practice – Observations and Questions Arising

The planning and creation of the HuComTech Multimodal Corpus is the latest sub-project of the HuComTech (Human-Computer Interaction Technologies) Research Team of Debrecen, Hungary. The sub-project is part of a project conducting research in multimodal human-machine communication. Currently the annotation of the corpus is underway, the questions arising during the annotation process are important from a linguistic point of view as well. Marking and differentiating the units of spontaneous speech and marking of disfluencies typical of spontaneous speech as well as structural and discourse characteristics have required to discuss the following problems: problems of segment boundary placement, differentiating elements with a pause-filling function, differentiating embeddings and insertions, differentiating iteration and restarting, usage of label hesitation and label backchannel in ambiguous cases. The present paper analyzes these questions and suggests solutions by describing examples from the corpus.