some foundational linguistic elements for qa systems: an application to e-government services farida...
TRANSCRIPT
Some Foundational Linguistic Elements for QA
Systems: an Application
to E-government Services
Farida Aouladomar
JURIX 05
9/11/05
Objective problematics of procedural texts for
answering procedural questions on the web governments increasingly offer online services
to their citizens. ex: How can I get the french citizenship?
Semantics and structure of procedural text answers to procedural questions are structures where to unify questions and parts of the
procedural texts give a complete and appropriate response
Advanced Question – Answering
Framework
QA Systems for factoid questions : InferenceWEB (McGuinness, 04), JAVELIN (Nyberg and al, 03), Mulder (Kwok and al, 00), WEBCOOP (Benamara, 04)
Non-factoid questions: comparison, procedural, opinion, causal, etc.
How-questions within a cooperative environment
How to get a passport ?
Response
Question analysis
Procedural Question?
Search engine
n texts
Filtering 1
Procedural texts ?
Annotation
Grammar
Filtering 2
Unification
Overview
Procedural questions
Grammar of procedural discourse - syntactic structure
- linguistic marks
Questionability and responses
Procedural questions (1)
Procedural questions = questions
introduced by « comment » - Comment vas-tu ?
- Comment dit-on maison en espagnol ?
- Comment est mort John?
- Comment on mange le couscous au Maroc ?
- Comment payer mon billet d’avion ?
- Comment changer une roue ?
- Comment créer une entreprise ?
Procedural questions (2)
Other forms of procedural questions :
– Forms in : « que faire … », « quel + être + proposition » …
Ex: que faire pour obtenir un visa ?, quelles sont les démarches à effectuer pour obtenir un visa pour l’Inde?quelle est la procédure pour obtenir la nationalité belge ?
– The elliptical use of « comment » : key words
Ex: trouver un avocat, déposer une plainte, inscription sur liste électorale …
– Lexical inference : Ex : j’ai perdu mon passeport
Questions expecting an answer of procedural type :
Ex: est-il possible d’avoir la double nationalité ?
Procedural questions (3)
Overview
Procedural questions
Grammar of procedural discourse
- syntactic structure
- linguistic marks
Questionability and responses
Sequences of instructions operating over a specific set of entities in order to reach a goal.
Goals & subgoals = skeletal structure of procedural texts
Group of texts : cooking receipes, maintenance manuals, assembly notices, directions for use, teaching texts, medical notices, social behavior recommendations, advice texts, savoir-faire guides, itinerary guides, architectory plans, musical scales, legislation, court decisions,legal guidelines & procedures, etc.
Corpus : identify textual elements of procedural texts for the answer process.
Procedural texts
Discursive Structure - Corpus (1) Structure analysis of procedural texts :
- Corpora analysis : method based on “how” queries on the web and on manual enrichment
Questions classificationQuestions classification Nb of procedural texts Nb of procedural texts from queries inventoryfrom queries inventory
Nb of added Nb of added procedural textsprocedural texts
Total number of Total number of texts in our corpustexts in our corpus
Communication / Communication / advices / advices / e-governmente-government
48 0 48
Technical domain Technical domain computer science computer science assembly textsassembly texts
30 20 50
HealthHealth 3 6 9
ReceipesReceipes 0 10 10
RulesRules 0 7 7
totaltotal 81 43 124
Discursive StructureGrammar (2)
Text → title, (summary), (warning)+, (pre-requisites)+, (picture) + < objective.
Summary → title+.
Pre-requisites → list of objects, (instruction sequences).
Instruction sequences → instseq < {connector} < instruction sequences / instseq.
Objective → {goal}< (warning), (picture), (pre-requisites), instruction sequences+ / objective
Grammar (3)
Imperative linear sequence → instruction < {temporal mark}, imperative linear sequence / instruction.
Optional sequence → optionality expression, imperative linear sequence.
Instruction sequence → imperative linear sequence / optional sequence / alternative sequence / imperative co-temporal sequence.
Imperative co-temporal sequence → imperative linear sequence < co-temporal mark < imperative co-temporal sequence / instruction.
Instruction → (iterative expression), action, (goal) (argument)+, (reference), (picture)+, (warning)
Alternative sequence → (conditional expression), (argument), imperative linear sequence, (alternative-opposition mark) < instseq / (conditional expression, instseq).
Grammar (3)
summary
title
Texte → title, (summary), (warning)+, (pre-requisites)+, (picture) + < objective.
warning
Objective
Objective → {goal}< (warning)+, (picture)+, (pre-requisites), instruction sequences+ / objective
goal
warning
Instruction sequences
Pre-requisites
Imperative linear sequence → instruction < (temporal mark) < imperative linear sequence / instruction
La premiere étape consiste à ouvrir entièrement le boîtier,
puis de le placer à plat sur une surface large où vous aurez suffisamment de place pour travailler confortablement,
et enfin retirer tous les caches en plastiques des baies à l’avant du PC
The first stage consists in fully-opening the box,
then place it on a large surface where you will have sufficient place to
work comfortably,
and finally to withdraw the plastic protections on the PC front side.
Instruction → (iterative expression), action, (goal) (argument)+, (reference), (picture)+, (warning)
La première étape consiste à ouvrir entièrement le boîtier,
puis de le placer à plat sur une surface large où vous aurez suffisamment de place pour travailler confortablement,
et enfin retirer tous les caches en plastiques des baies à l’avant du PC
La premiere étape consiste à ouvrir entièrement le boîtier,
puis de le placer à plat sur une surface large où vous aurez suffisamment de place pour travailler confortablement,
et enfin retirer tous les caches en plastiques des baies à l’avant du PC
ArgumentTemporal marks Instructions
Texte → title, (summary), (warning)+, (pre-requisites)+, (picture) + < objective.
title
Pre-requisites
Objective
annotation < objective> <goal> Postez le formulaire et les documents au Centre
de traitement des demandes. <\goal>
< instseq> < imper_linear> <temp_mark> Après <\temp_mark> <instr> avoir rempli le formulaire de demande, <\instr> <instr> vous devez le poster, dans l’enveloppe-réponse fournie, à l’adresse suivante :
Centre de traitement des demandesCitoyenneté et Immigration CanadaC.P. 7000Sydney (Nouvelle-Écosse) B1P 6V6 <\instr> < imper_linear> < \instseq>
<warning> N’oubliez pas : de signer et de dater le formulaire ainsi que de signer vos photos; d’inclure le reçu du paiement (formulaire IMM 5401); de mettre votre demande dans l’enveloppe; de mettre les photos dans l’enveloppe; d’inclure les photocopies de tous les documents requis. <warning>
<\objective>
< instseq> < imper_linear> <temp_mark> Après <\temp_mark> <instr> avoir rempli le
formulaire de demande, <\instr> <instr> vous devez le poster, dans l’enveloppe-réponse fournie, à l’adresse suivante :
Centre de traitement des demandesCitoyenneté et Immigration CanadaC.P. 7000
Sydney (Nouvelle-Écosse) B1P 6V6 <\instr> <\imper_linear> < \instseq>
<warning> N’oubliez pas :<instr> de signer et de dater le formulaire ainsi que de signer vos photos <\instr>;
<instr> d’inclure le reçu du paiement (formulaire IMM 5401) <\instr> ; <instr> de mettre votre demande dans l’enveloppe <\instr>;
<instr> de mettre les photos dans l’enveloppe <\instr>;
<instr> d’inclure les photocopies de tous les documents requis. <\instr> <warning>
The grammar is :
-- too rigid, and not sufficient to describe all facets of procedural texts
-- not very constrained
-- not explicative and predictive enough
A new orientation of the discursive structure of procedural texts : principles and norms
Grammar limits
Overview
Procedural questions
Grammar of procedural discourse
- syntactic structure
- linguistic marks
Questionability and responses
Linguistic marks (1)
Discursive marks : allow for the identification of the elements of the grammar
-- classical temporal marks (precedence, overlap, inclusion, etc.)-- restrictions, conditions, alternatives, comparisons, etc. -- causal marks (identification of objectives, goals, warnings, preventions, consequences, etc.)
-- typoraphic criterion:
instruction localization (1)
-- morphological criterion (ex: imperatives)
instruction localization (2)
-- semantic criterion: action verbs
instruction localization (3)
Overview
Procedural questions
Grammar of procedural discourse
- syntactic structure
- linguistic marks
Questionability and responses
How to download music ?
Response
Question analysis
Procedural Question?
Search engine
n texts
Filtering 1
Procedural texts ?
Annotation
Grammar
Filtering 2
Unification
QUESTIONABILITY
Questionability (1)
The ability or the relevance of any text to respond to How-questions.
Stage 1: determine the « CATEG » rate
-- measure the procedural nature of a text
Stage 2: determine the « QUEST » rate
-- measure of zones which are qualified for potential question unification
Questionability (2)
Stage 1 : « CATEG » => 3 « surface criteria »
- typographic forms (TF)- morpho-syntactic marks (MSM)
- articulatory marks (AM) : (temporal, argumentative, etc.)
average frequency is computed for each criteria for all texts (noted as TFaverage, MSMaverage,
AMaverage)
Questionability (3)
For text i, we Define its CATEG rate:
Selection of the « best » texts => annotation
Questionability (4)
Stage 2 : « Quest » => evaluate the number of areas which can potentially match with How-questions
4 areas leading to 4 criteria : - number of titles (TIT)- action verbs (AV)- number of goals (GOA)- manners (MAN)
Questionability (5)
For text i, we define its QUEST rate :
Question Unification
Responses
3 tasks :
-- selecting procedural texts which have the best questionabilty rate
-- matching the question body with « questionable zones »
-- extracting the relevant portion of the text and returning it to the user in a user-friendly way
To Conclude … Still experimental work
- ongoing : design of a system that annotates texts and evaluates the questionability rate
Perspectives :- validation / evaluation
- unification give a complete answer to the user
- respond cooperatively to procedural questions
Work more specifically on the structure of legal / administrative procedural texts