An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 8

consequent manner: it becomes aggressive if the anger value passes a certain
threshold. The rules or heuristics, on the contrary from ELIZA's rules, are
based both on the user's utterances and on the variables of state. PARRY
marks an evolution of MMD systems, with the technical means of the time:
the program, written in a variant of the Lisp language, takes 35 Kb of which
14 Kb belong to the database.
The 1970s were the time of the first (written) understanding systems, with
significant improvements in NLP, especially in syntactic and semantic
analyses and, thus, the first true systems of MMD written that model a field of
knowledge, know how to interpret an utterance in this field, and start to
manage a structure dialogue. This progress follows a few landmark works in
linguistics and computer linguistics, especially B.J. Grosz and then C.L.
Sidner; as Jurafsky and Martin [JUR 09, p. 892] underline it. That
corresponds to the first path mentioned on page 6, with two key systems,
SHRDLU and genial understanding system (GUS). In parallel, the speech
recognition system path is also progressing strongly, especially with systems
developed within the American Advanced Research Projects Agency (ARPA)
projects: Harpy, Hearsay, Hwim. We thus go from the recognition of isolated
words, which is not at all adapted to MMD, to the recognition of continuous,
and eventually multi-locutor words, with concerns which start to reach those
of MMD, for example the question of software architecture to get various
sources of knowledge communicating inside systems, see Pierrel's [PIE 87]
historical outline. We will return to this in section 1.1.2 with the first oral
MMD systems.
The SHRDLU 2 system [WIN 72] gives a new boost to written MMD by
showing the deeper understanding and dialogue possibilities as soon as you
limit yourself to a clearly limited and modeled task. This time, let us forget
about the Turing test and turn to targeted applications: the task consists of
displacing geometrical objects (cubes, cones and pyramids) with a machine.
It involves the display of a scenario on a screen, with a representation of the
system itself with a kind of robot arm manipulating objects. The user creates
utterances such as “pick up a green block” or “find a block that is taller than
the one you are holding and put it into the box”, and the system carries out
2 The name comes from the sequence of lettersETAOINSHRDLUthat is, in decreasing
order, the sequence of letters most often used in English, in the way they are vertically shown
in the middle of some printing machine keyboards.