Semantic Filtering of Textual Requirements Descriptions Jorge García-Flores LaLICC Université de Paris Sorbonne.

Semantic Filtering of Textual Requirements Descriptions

Jorge García-FloresLaLICCUniversité de Paris Sorbonne

73% of documents availiable for requirements analysis are written in natural language.

Natural Language Processing is a useful tool to handle with this mass of documents

[Mich et al., 2003]

Use cases Scenarios

User storiesElicitation documents

Rough sketches

Domain manuals

Motivation

To support requirements analysis and validation of large Textual Requirements Descriptions by filtering TRD according to semantic criteria

To apply a linguistic technique (the Contextual Exploration Method) to the Requirements Engineering process.

Goal

Existing approaches

Linguistic-based NLP technique Oriented to semantic filtering of large

document by shallow parsing Rule based structure It takes into account structural (title,

subtitle, section) and typhographical elements

It has been used for automatic summarisation and large text filtering

Contextual Exploration

1. Search of linguistic markers in a text

2. Search of linguistic clues in the marker’s context

3. If the appropriate clues are found, a semantic value is assigned

CE rule

CE rule (was he captured?)

In spite of all precautions, he was captured the day after

Without all precautions, he would have been captured the day after

Accomplished

Not_accomplished

LinguisticAdministrator

SystemLinguistic Repository

Text segments.Tasks

Rules, markers,clues

ContextualExploration

Engine

Dedicatedagent 1

Dedicatedagent 2

Dedicatedagent 3

Linguisticressources

configuration

SemanticFiltering

(CE)

Concept relationships

Tense & aspect

Control

Causality

Browsing

Semanticvalue

processing

Schemageneration

!

partial views

conflictingviews

conceptualschema

Semantic filtering of TRD

GAC-based semantic viewpoints

I. Concepts relationshipsII. Aspecto-temporal

organization III. ControlIV. Causality

[Desclés, 1990]

“When the start button is pressed, if there is an original in the feed slot, the photocopier makes N copies of it, and places them in the output tray. N is the number currently registering in the count display. If the start button is pressed while photocopying is in progress, it has no effect. The number N in the count display updates in response to button pressed according to the state table”

[Kovitz, 1998]

An exemple

“When the start button is pressed, if there is an original in the feed slot, the photocopier makes N copies of it, and places them in the output tray. N is the number currently registering in the count display. If the start button is pressed while photocopying is in progress, it has no effect. The number N in the count display updates in response to button pressed according to the state table”

equality

Concept relationships

“When the start button is pressed, if there is an original in the feed slot, the photocopier makes N copies of it, and places them in the output tray. N is the number currently registering in the count display. If the start button is pressed while photocopying is in progress, it has no effect. The number N in the count display updates in response to button pressed according to the state table”

event

process

Tense and aspect

“When the start button is pressed, if there is an original in the feed slot, the photocopier makes N copies of it, and places them in the output tray. N is the number currently registering in the count display. If the start button is pressed while photocopying is in progress, it has no effect. The number N in the count display updates in response to button pressed according to the state table”

environment controlled

machine controlled

Control

“When the start button is pressed, if there is an original in the feed slot, the photocopier makes N copies of it, and places them in the output tray. N is the number currently registering in the count display. If the start button is pressed while photocopying is in progress, it has no effect. The number N in the count display updates in response to button pressed according to the state table”

Causality

« Only the system can trigger a premium-collection event. »

« The system can prevent a premium-collection event but only an agent can cause it. »

page 60

page 234

CONTR

CAUSE

CONTR

Conflicting views

TRDViewpointbrowser

ContextualExploration

system

SV- basedrules

Linguistic Repository

SemanticFilteringbroker

Semanticvalues

analyzer

Rulesconfiguration

tool

TRD

Filtered textSemantic values

Rule

sSchema

generationtool

Architecture

Evaluation of the presence of current CE rules, markers and clues on industrial requirements documents.

Definition of a declarative language for semantic-value based rules in a way that could allow interoperability between viewpoints

Current work

Gracias!