|
|
|
|
|
•The text planner creates text plans. A
text plan is a tree structure which organizes the content of the A-box in a
linear fashion. The leaves of the text plan identify the particular content
(given in the A-box) that is to realised “there”. The creation of a text plan
for a particular text type is implemented as a multilingual region in KPML.
As there are several text types, we thus have several regions. Text plans may
plan a large variety of phenomena – discourse/rhetorical relations
(RST-purpose, means, logical-condition), conjunction and disjunction,
different types of temporal sequence marking, local stylistic variation
(running/listing text, different aspect), semantic grouping (whereby a semantic
lattice acts as a conduit between an A-box and the text planner). The
text plan as such is language-independent, however we are able to
include language-specific realisation constraints to guide lexico-grammatical
generation. Such constraints are for example used to realise global stylistic
variation (personal vs. impersonal style). We currently have regions for full
procedural instructions, quick references, tables of content, functional
descriptions, and overviews. All of these text types can be generated from
the ‘standard’ A-boxes!
|
|
|
|
The text planner
is “incremental”, in that it can be called not just once an A-box is
finished, but throughout the construction process. This facilitates
WYSIWYM-like construction of content, following a
construct-generate-evaluate-reconstruct/extend cycle.
|
|
|
|
The sentence
planner interprets the structure of a text plan, and creates sentence plans
in the SPL language. The sentence planner is implemented in LISP. What it
does is that it follows the structural relations between content as indicated
in the text plan to aggregate content into sentences – the sentence planner
does not make any such decisions of its own. The advantage here is
that we obtain a very flexible way in which we can develop text grammars:
namely, by using KPML. The only “changes” we need to make to the sentence
planner is to add guidelines how to interpret node-labels appearing in the
text plan.
|
|
|
|
A notable
feature of the sentence planner is that it makes use of a (rudimentary)
discourse model to include information structure in the sentence plans. The
discourse model keeps track of salient items in the text plan (new/reused identifiers). Whenever a
sentence plan specifies the realisation of an already “given” item, then that
content is planned as part of the sentence’s topic. The
lexico-grammars realise information structure by means of variation in word
order. Another feature of the sentence planner is that it creates sentence
plans that have –where needed- language-specific complex semantics to realise
particular DM concepts. The paper lists a few examples.
|
|
The TSM is
implemented as a KPML module: Bridging the gap – it is very easy to impose
constraints on lexico-grammatical realisation.
|
|
|
|
Further
references: TSD papers, ESSLLI/book
|
|
|
The AGILE TSM consists of a text planner and a
sentence planner
