Information Technology Reference
In-Depth Information
Assignment of segments to categories on the basis of grammatical information:
Identification of the four segment classes is possible on the basis of the Part-of-Speech
(POS) tags: A POS tagger decides, for every word, if this word is a noun, verb, adjec-
tive, ...The applied tagger has the precision of about 97% [8] which makes it unlikely
to become an error source. Furthermore, we assume that the names of the automaton
states are extracted from the specification before the actual text-to-automaton transla-
tion, cf. [6]. The assignment of the sentence segment to one of the four classes takes
place in the following way:
-
If the sentence segment does not contain any reference to a state, it is marked as
“irrelevant”. This holds, for example, for the first segment in Table 2.
-
If the sentence segment contains a reference to a state, but first occurrence of the
state is not preceded by a verb, this segment is marked as “context setting”. For
example, in Table 1, the header (“normal mode”) and the first sentence set the
context for the translation of the following sentences.
-
Otherwise, the sentence segment is marked as “state transition”.
Here it is important to emphasize that in the first phase no sentence segment is marked
as “transition condition”.
Re-assignment of segments to categories, by using context information:
To take
context into account, it is necessary to revise the “context setting”-marks first. Here, the
following heuristics is applied: If, for a given sentence, any of its segments is marked as
“state transition”, then all segments marked as “context setting” are relabeled to “state
transition”. This compensates for potentially missing verbs in some sentence segments.
In the case of the example shown in Table 2, it marks the fourth segment as “state
transition” and leaves the other marks unchanged.
When the marking of segments as “state transition” is finished, it is possible to iden-
tify transition conditions:
-
If a sentence segment is marked as “irrelevant” and directly precedes a segment
marked as “state transition”, then the former segment is relabeled to “transition
condition” (e.g., the first segment of the example in Table 2).
-
If a sentence segment is marked as “irrelevant” and directly follows a segment
marked as “state transition”, then the former segment is relabeled to “transition
condition”. This allows to treat conditions like “
some transition
if
some
condition
”.
When this relabeling process is finished, transitions are created from the sentence seg-
ments marked as “state transition”. Transition conditions, as well as source and target
state of transitions, are inferred from the adjacent “context setting” and “transition con-
dition” segments.
The presented approach is domain-independent and only relies on a special writing
style with guidelines for industrial requirements specifications: The complete set of
system states is known, sentences describing state transitions contain a reference to the
target state, and context setting is stated explicitly (for details, see [6]).
