Hervé BLANCHON, Laurel FAIS
Pilot Experiment on the
Understandability of Interactive
Disambiguation Dialogues
Abstract:We report on a pilot experiment that was carried out at ATR-ITL on the topic of interactive
disambiguation and more precisely, on the understandability of interactive disambiguation
questions.
Two classes of questions (human-like and machine-like) were proposed using a textual modality.
The human-like questions were worded in natural, conversational language, and the machine-like
questions were the ones that we would have been able to produce using the disambiguation
methodology proposed in [1-3] The answers to the questions were of two kinds, "easy"
(corresponding to the most frequent or most natural interpretation of a given sequence of words)
and "hard" (corresponding to a very infrequent or unnatural interpretation of a given sequence of
words). We had two groups of subjects, each subject participating in only one setting.
A text containing 35 questions (21 hard, and 14 easy) was read aloud by the subjects, who then
answered the questions. The analysis of the results shows that the experiment was not well
designed. The text was too difficult and unnatural to enable the subjects to focus their attention
correctly; they performed poorly with each one of the two sets of questions.