Information Technology Reference
In-Depth Information
3
Feedback Generation Instead of Inference Assumptions
The translation approaches presented in Section 2 use different inference rules in order
to complete information not explicitly stated in the text. The main idea of feedback gen-
eration is to turn off the inference, and, whenever the inference would become neces-
sary, to generate feedback questions addressing the missing information. For MSCs, this
missing information can be the unspecified message receiver or an unspecified interme-
diate message that is inferred by the means of the stack model. Feedback generation
for MSCs is presented in Section 3.1. For automata, the missing information is always
the source state of a state transition (due to peculiarities of the algorithm presented in
Section 2.2). Feedback generation for automata is presented in Section 3.2.
In general, the generated questions can refer either to the specification text or to the
extracted model. Reference to the model means that, in addition to the actual models,
special markers are generated that pinpoint the problematic model elements. Reference
to the text means that hints in natural language are generated, in the following form:
- Problematic sentence:
sentence cited
- Detected problem:
problem description
Both for automata and for MSCs, we implemented the generation of both representa-
tions.
3.1
Feedback Generation for MSCs
The algorithm for MSC generation presented in Section 2.1 can infer unspecified mes-
sage receivers and whole missing messages. Correspondingly, the feedback questions
that are generated for MSCs, address (1) unspecified message receivers or (2) messages
that are necessary from the point of view of a continuous message flow, but not explic-
itly specified in the document.
Every type of the feedback question (addressing the missing receiver or the missing
messages) can refer both to the model and to the specification text. Reference to the text
means that we generate questions or hints in natural language, addressing the detected
problem. Reference to the model means that we use special markers in the generated
MSCs in order to address the detected problems.
Missing message receivers result from active sentences only, as passive sentences
are translated to MSC assertions, cf. [4]. Furthermore, an active sentence needs a gram-
matical object to be a candidate for an MSC message. Nevertheless, if the sentence
object is not contained in the set of potential MSC actors, inference rules sketched in
Section 2.1 must be applied. Thus, in order to address missing message receivers, we
generate schematic hints referring to the specification text, structured as follows:
Problem:
Problematic active sentence: object
sentence cited
sentence object
is not con-
tained in the list of actors. Message receiver cannot be identified.
The inference of missing messages becomes necessary when the sender (basically,
grammatical subject) of the sentence under consideration does not coincide with the
receiver of the last message. In this case we generate coherence questions:
