Introduction: How to Build a Talking Robot - Computational Linguistics and Talking Robots - page 6

A crucial part of a declarative specification is the choice of the basic unit -
analogous to that of other sciences founded on basic units such as the cell in
biology or the atom in chemistry and physics. The basic unit of DBS is the pro-
plet , defined as a flat feature structure 12 with a finite set of ordered attributes
and double-ended queues as values. Compared to the recursive feature struc-
tures used in Nativism (with unordered attributes, but an order of embedding),
proplets have the following advantages:
1.3.1 A DVANTAGES OF PROPLETS
1. Flat ordered feature structures are easier to read and computationally more
efficient than recursive feature structures with unordered attributes.
2. Flat ordered feature structures provide for easy schema derivation and for
easy pattern matching.
3. The combination of a proplet's core and prn value provides a natural pri-
mary key for storage in and retrieval from memory.
4. Coding the semantic relations between proplets as addresses (Sect. 4.4)
makes proplets order-free and therefore amenable to the needs of one's
database.
5. The semantic relations between proplets enable time-linear navigation
along those relations, reintroducing order and serving as the selective acti-
vation of content, as needed in language production and inferencing.
In summary, a proplet is defined as a list of features (internal order) and each
feature is defined as an attribute-value pair ( avp ). The proplets representing a
complex content, in contrast, are a set (no external order, e.g., 3.2.1).
The data structure of proplets has been used in the LA-hear, LA-think, and
LA-speak grammars defined in FoCL'99 and NLC'06. They map sequences of
natural language surfaces into sets of proplets (hear mode), and sets of proplets
into corresponding sequences of natural language surfaces (speak mode).
Proplets turn out to be versatile in that they maintain their format and their
formal properties in a multitude of different functions. Consider the following
examples of a language proplet for German, a language proplet for French, a
content proplet, and a pattern proplet:
that competence grammars are not intended to model the language processing by the speaker-hearer.
Therefore it is hardly surprising that the functional flow of competence grammars (NLC'06, 3.4.5) is
incompatible with that of a talking robot. Also, competence grammars are cumbersome to program
because they fail to be type-transparent (Berwick and Weinberg 1984; cf. FoCL'99, Sect. 9.3).
The declarative specification of DBS, in contrast, is an entirely computational theory. It aims at
completeness of function and of data coverage as well as low complexity because the system must be
efficient enough to run in real-time.