Conclusion - Computational Linguistics and Talking Robots - page 235

The flexibility required by the human user depends in part on what the part-
ners in discourse accept as communication. For example, high brow journals
will accept only papers which exhibit the correct use of the current buzz words
in their domain, some bureaucrats are interested only in a small range of topics,
such as last name, first name, date of birth, and place of birth, the customers
in a bar, who will accept a stranger only if (s)he is a certain type, and so on. In
linguistics, these restrictions of a domain to a certain, well-defined protocol are
called register (Halliday and Hasan 1976). Register is important for commu-
nication, natural or artificial, because it defines the channel of communication
from a social point of view.
In this sense, the construction of a talking robot must solve the task of
register adaptation . The system should be able to smoothly agree with the
partner in discourse on a certain level of abstraction, to switch the level of
abstraction up or down, accompanied by adjustments of dialect and of into-
nation, to select between a declarative, interrogative, or imperative sentential
mood, to present content in a certain way, etc. This is the most fertile field of
Conversation Analysis, founded by Sacks and Schegloff (Schegloff 2007). 5
The DBS robot requires a computational model of the motivational struc-
tures behind the actions of the partners in discourse and of the strategies guid-
ing these actions. For this, the Schegloff corpus provides many authentic ex-
amples. A DBS interpretation of the Schegloff corpus would have to translate
such founding notions as “pre,” “post,” “pre-pre,” etc., (time-linear!) into DBS
inferences with certain goals.
To overcome the inflexibility of current systems. Mohammadzadeh et al.
(2005) propose “template guided association,” aimed at XML. The DBS ap-
proach is similar: the templates are the schemata built from pattern proplets. In
addition to a highly effective primary key, DBS provides the option to search
for continuation values, morphosyntactic categories, base forms, matching in-
ferences, memorable outcomes, n-grams, frequencies, and so on.
The conceptual backbone of DBS is storing proplets in the order of their
arrival 6 in combination with the “no-change” rule of a content-addressable
memory. Because the processing of content never modifies what is stored in
memory, there is a strict separation between the storage of content and its
processing. The storage operations, like the inferences, always write to the now
5 Schegloff's examples are interesting and carefully analyzed, but a computational implementation
was not one of Schegloff's goals. The observable facts are well documented, but the difficulties of
interpretation or of choosing between several possible interpretations are pointed out repeatedly.
6 In DBS, time is represented solely by the proplets' relative order of arrival. There are three possibili-
ties: proplet A is earlier than than proplet B, proplet A is later than proplet B, or proplets A and B are
simultaneous. This order is reflected by the prn values of the proplets.