Conclusion - Computational Linguistics and Talking Robots - page 247

language, the LA-hear parser will at first handle only a few constructions. As
the language is being studied, more and more constructions (like infinitives,
prepositional phrases, relative clauses, etc.) are added to the grammar, tested,
and revised. When the LA-hear parser encounters input it cannot yet handle,
the passage may be traversed at a lower level of detail until proper parsing can
resume (robustness). For this, LA-grammar is especially suitable because it
computes possible continuations in a time-linear derivation order.
Expanding syntactic-semantic parsing in the agent's hear mode is more de-
manding than automatic word form recognition. This effort should not go un-
rewarded from the application side, however. The coding of functor-argument
and coordination extends recall and precision from lexically analyzed word
forms to phrases and clauses, and from there to sentences, paragraphs, and
text. Technically, this amounts to an extension from matching lexically ana-
lyzed content words stored within token lines in the Word Bank, to matching
semantic relations between content words defined across token lines. 30
The think mode is a third area for extending the data coverage. The agent's
think mode combines two mechanisms, LA-think and inferencing. The basic
mechanism of LA-think is selective activation by navigating along the seman-
tic relations in a Word Bank. 31 The navigation is used to activate and report

self-contained content.
Inferences are used for deriving the different perspectives of the speaker and
the hearer on content, 32 and to compute blueprints for action, including lan-
guage action. Together with current and stored data, LA-think and inferencing
constitute the agent's autonomous control, which has many practical applica-
tions, with and without language.
Finally, consider LA-speak. It takes content as input and produces corre-
sponding surfaces as output. If the content has already been serialized by the
navigation along the semantic relations in the Word Bank, the task of LA-
speak is confined to adjusting to the word order of the language and to pro-
viding proper lexicalization with proper perspective (e.g., tense) and proper
morphosyntactic adjustments (e.g., agreement).

30 In addition, the user may load the proplets in a proprietary database with additional attributes and val-
ues as needed for the application. One such application of LA-morph and LA-hear is speech recogni-
tion; it could well benefit from the search space reduction resulting from an LA-hear parser computing
possible continuations (Sect. 2.4).
31 A Word Bank may be viewed as a syntactic-semantic network. For some questions and results of
linguistic networks, see Liu (2011), Solé et al. (2010), Sowa (1987/1992), Brachman (1979), and
others.
32 The interaction between LA-think, LA-speak, and inferencing is shown in Chap. 10 with an example
of dialogue.