Design
Implementation
| •Re-use/adaptation of existing resources: In
designing (and building) the system we have made use of, and adapted where
necessary, various resources already in existence: |
||
| •Core modules are all implemented in the
Komet-Penman Multilingual system |
||
| •The Penman Upper Model was adopted, and has
been extended with a domain model relevant to the CAD/CAM domain |
||
| •The domain model was inspired by the domain
model for the Drafter system which also dealt with instructional texts,
though the AGILE DM presents a significant generalization over the former in
that e.g. semantic structures modeling instructions can be of arbitrary
recursive depth |
||
| •Already existing lexical resources and
morphological modules available for the languages under consideration have
been (successfully) interfaced with KPML |
||
| •Finally, because no generation grammars (or
large-coverage formal grammars) existed prior to this project, a grammar for
English (NIGEL) was re-used to build them. Even though Slavic languages
differ significantly from English, the systemic/functional organization of
NIGEL made adaptation feasible. |
||
| •Combination of two methods of resource
development |
||
| •The system-oriented approach, conceiving of
language system as a whole, which is a perspective strongly supported
by the KPML environment |
||
| •The instance-oriented approach, guided by a
detailed analysis of the register(s) for which the grammar were to be
developed. This approach was particularly important to the project given the
overall goal of being able to generate texts belonging to rather diverse (I
would just say: several different)
text types and styles: personal vs. impersonal style, full procedural
instructions versus functional descriptions, etc. |
||
| •Cross-linguistic resource sharing |
||
| •The systemic/functional approach, arranging a
grammar “vertically” with modules describing cross-linguistic commonalities
high up in the system such that the grammar only gets more language-specific
towards the “leaves”, greatly facilitates cross-linguistic resource sharing. |
||
| Integrated
system, consisting of |
||
| •Multilingual lexico-grammatical resources
(including morphological and lexical resources) |
||
| •Multilingual text structuring module (text- and
sentence planner) |
||
| •Graphical user interface for content
specification (A-box construction), selection of various generation options
(styles, text types, etc.) |
||
| •Coupling to an internet browser (IE) in which
the generated texts are displayed (including layout / HTML) |
||
| •Multilingual text- and sentence-planning of a
variety of text types |
||
| •Text planning of a variety of text types,
including for example: full
procedural instructions (corresponding to the texts as found in the user
manual), whose planning includes discerning various rhetorical relations,
temporal sequencing, “syntactic” aggregation like and-/or-coordination; Overviews (summarizing the content of
several FPIs), whose planning includes (dynamic) semantic grouping, stylistic
variation to indicate change in content; Fd’s …I’d at least mention them;
they describe individual commands etc. |
||
| •Sentence planning to generate sentence plans in
SPL, including for example: Language-specific lexical realisation of DM
concept; Interpretation of the text
plan to build e.g. complex clauses (aggregation); Imposition of information
structure based on a discourse model built over the text plan (here you’ll have to be more specific, what we mean) |
||
| •Multilingual grammar |
||