3 What are spoken language data?

Two epistemological domains:

SIGNALS: recorded spoken data

SYMBOLS: interpreted spoken data

The SIGNAL-SYMBOL barrier:

HUMAN: categorial, interpretative perception, world knowledge

MACHINE: stochastic segmentation, classification, top-down prediction

Pre-recording, recording post-recording requirements:

1. Corpus design: specification of speakers, scenario
2. Corpus recording: studio quality
3. Corpus processing: physical and linguistic characterisation:
   - Physical: signal properties, speaker characteristics
   - Linguistic: transcription, annotation, lexicon ...