Corpus Construction
LABELING EFFORT
Hand correction of automatically (HMM)
pre-aligned labels by 7 naive labelers trained
for this specific task
Optimum labeling speed reached after
40 hours of transcription
Top speed was 1 +/ 0.2 words per minute
Version management using CVS over internet
Total amount of labeling work:
50,000 words and 200,000 segment boundaries
Total time of labeling:
1000 hours (manual labeling)
6 person-months (staff-time overhead etc.)
Average labeling speed:
0.84 words / minute
3.3 boundaries / minute
Without pre-alignment, labeling speed
approximately halves
Monetary cost of alignment:
(excluding VAT, automatic and manual alignment combined)
DFl 74,000 Total (= 33,597 Euro)
DFl 1.40 / word (= 0.65 Euro / word)
DFl 0.37 / boundary (= 0.17 Euro / boundary)