EA-Miner: Towards Automation in Aspect-Oriented Requirements Engineering - Transactions on Aspect-Oriented Software Development - page 9

(4) Validating requirements and resolving conflicts: This step represents the
validation of the specified requirements and also conflict detection and resolution.
EA-Miner's output of step 3 can be easily used as input by other tools (e.g., Arcade
[28]) that perform conflict identification. For example [16, 17] shows how the early
aspects can contribute to each other (e.g., security and response time can both
crosscut the same requirements of an ATM viewpoint and can contribute negatively
to each other).
One important point to mention is that the previous activities are not necessarily
conducted in a strict sequential order but in an iterative and incremental fashion
providing forward and backward feedback cycles. The next section explains how NLP
techniques can be used to address mining of concepts (activity 2) in different AORE
approaches as well as structuring and filtering capabilities (activity 3).

3 Utilizing NLP Techniques for Automation

The cornerstone of EA-Miner's model automation are the natural language processing
features provided by the WMATRIX NLP tool suite which have been shown to be
effective in early phase requirements engineering [11, 28, 29]. WMATRIX
implements NLP techniques such as frequency analysis, part-of-speech (with a
precision of 97%) and semantic tagging (with a precision of 91%) that provide
relevant information about the properties and semantics of a text in natural language.
Frequency analysis shows statistical data about frequencies of words that help to find
out which ones are more significant in the text.
WMATRIX takes a corpus-based NLP approach. Corpus Linguistics [30] can be
understood as the study of language based on “real life” language use. A corpus is a
collection of texts from different sources (e.g., newspapers, magazines, books,
journals) that can be collected over several years and made available for researchers.
For example, the British national corpus (BNC) [31], on which WMATRIX draws, is
a 100 million word reference collection of samples of written and spoken English
from a wide range of sources.
Part-of-speech (POS) tagging [11, 28] assigns to each word its grammatical
function (part-of-speech) such as singular common noun, comparative adjective,
infinitive verb and other categories such as the ones in Table 1. The tagging process
in WMATRIX is based on a language model derived from the large reference corpus
and uses surrounding context to decide the most likely tag for each word.

Table 1. Examples of POS and semantic tags from [11, 28]

POS TAG WHAT IT REPRESENTS

NN1 singular common noun (e.g. topic, girl)

VVI infinitive (e.g. to give... It will work...)

SEM TAG WHAT IT REPRESENTS

M movement, location, travel and transport

S social actions, states and processes

M3 vehicles and transport on land

S7.4 permission