Human-robot speech interface understanding inexplicit utterances using vision