Man–Machine Dialogue Fields - Man-Machine Dialogue: Design and Challenges - page 37

any type of knowledge transmission, and the dialogue between an expert and
a learner is the very first corpus. Depending on how the dialogue goes, we can
distinguish active learning, during which the learner asks questions and in
some ways directs the dialogue according to his desires, and passive learning,
during which the expert directs the dialogue and transfers knowledge. The
language itself is the focus of learning, and the dialogue between an adult and
a child is a second corpus of study, as is the dialogue between an expert and a
learner trying to acquire a second language. [CLA 09] thus presents a list of
phenomena characterizing the dialogue between an adult and a child. First,
the child tends to mispronounce words, which leads the adult to
over-pronounce them. The child makes morphological mistakes, for example
when conjugating verbs, lexical mistakes and syntactic mistakes. This leads
the adult to correct him, either when the child has finished talking, or by
interrupting him. At the dialogue structure level, the main observation is that
the child does not follow the same structure that can be observed in a dialogue
between two adults: the child is often self-centered and tends to follow his
own trail of thought rather than interact with the adult. Moreover, the adult
tends to prefer closed questions, whose answers are yes or no, to open
questions, whose answers require expressing knowledge. He creates
repetitions and rephrasing, to increase the chances of information
transmission. He also adapts himself to the words that the child knows, so he
aligns himself with the child from a lexical point of view and follows a
chronological, or at least logical order, when telling a story, for example.
These learning situations and their characteristics can inspire MMD
system designers, without the system or the user being considered as a
learner. In situations where the system has to give the user an explanation, if
only to have him wait while carrying out the query, providing information in a
chronological manner can help the generation and in the same way,
generating closed questions can also do so. If the machine is able to learn
online, that is to enrich its linguistic abilities during the dialogue and through
it, we can plan for specific processes for the user's utterances with the aim of
correcting or of assessing the achieved progress.
However, human learning and machine learning are very different, and a
machine does not have to go through all the stages a child does. In the case of
language learning, if it is decided to provide the machine with this ability,
then the learning does not focus on all the dimensions of the language:
pronunciation and morphology are much less problematic than lexical
semantics or syntax. The MMD does not focus on simple mistakes, and there