Linguistic Realisation II

1
Linguistic Realisation II

• TILT 2003

2
Lecture Overview

• What is linguistic realisation?
• Processing tasks in NLG systems
• Modular structure of NLG systems
• Example input/output of realiser
• Subtasks in realisation
• Different ways of doing realisation
• Research issues in realisation
• Lab session extend code from Lecture 1

3
What is linguistic realisation?

• Depends on NLG system
• Can cover different subtasks lexicalisation,
  generating referring expressions, applying
  syntactic and morphological rules
• Can be done in different ways constructing from
  scratch, canned phrases and substructures
• First approximation linguistic realisation is
  deciding how to express a given meaning.

4
NLG Tasks

• Content determination decide what to say
  construct set of messages
• Discourse planning ordering, structuring
  concepts rhetorical relationships
• Sentence aggregation divide content into
  sentences construct sentence plans
• Lexicalisation map concepts and relations to
  lexemes
• Referring expression generation decide how to
  refer to objects in the messages
• Linguistic realisation put it all together in
  acceptable words and sentences

5
Modular structure of NLG systems (in theory!)
Content determination
TEXT PLANNER
Discourse planning
Sentence aggregation
SENTENCE PLANNER
Lexicalisation
Referring expressions
REALISER
Realisation
6
Conceptual grouping
Content determination
WHAT TO SAY
Discourse planning
Sentence aggregation
Lexicalisation
HOW TO SAY IT
Referring expressions
Realisation
7
In practice

• No two NLG systems have exactly the same modular
  structure
• Lexical and syntactic choice can be part of
  almost any module
• Many systems dont even have a strict separation
  between deciding what to say and how to say it

8
So what is realisation?

• Realisation is (for our purposes) deciding how to
  express a given meaning by
• Selecting suitable words to express concepts and
  relations
• Combining the selected lexemes in agreement with
  the given meaning and the languages sentence and
  word formation rules
• Noun phrases may be generated by separate module
  for generating referring expressions
• NB More often, realisation 2.

9
Input to realisation

• message-id msg02
• relation C_DEPARTURE
• departing-entity C_CALEDON-EXPRESS
• args departure-location C_ABERDEEN
• departure-time C_1000

10
Output from realisation

• Lexicalisation/GRE
• C_DEPARTURE depart
• C_CALEDONIAN-EXPRESS the Caledonian
  
• Express
• C_ABERDEEN Aberdeen
• C_1000 10am
• Syntax and morphology
• The Caledonian Express departs Aberdeen at
  10am.

11
Alternatively

• Lexicalisation/GRE
• C_DEPARTURE leave
• C_CALEDONIAN-EXPRESS it
• C_ABERDEEN here
• C_1000 in five minutes
• Syntax and morphology
• It leaves here in five minutes.

12
Subtasks in Realisation

• From thematic roles to syntactic roles John sees
  Mary, Mary is seen by John
• Decide between alternatives Give the book to
  him, Give him the book
• Ensure right order of constituents Him give the
  book
• Ensure syntactic agreement Mary are seen by
  John
• Ensure correct inflection Mary is seed
• Insert certain words Mary is seen by John

13
Different approaches to realisation

• Generation as reversed parsing use single
  grammar that defines the relation between meaning
  representations and surface realisations
• Grammar-based linguistic realisation use a
  linguistically motivated grammar of the language,
  often split realisation task in two
  microplanning and surface realisation
• Possibility of incorporating arbitrarily large
  chunks of canned text

14
Existing Realisers

• Reversed parsing ALE, ASTROGEN, Gemini
• Template-based realisers TG/2, YAG
• Grammar-based linguistic realisers PENMAN, KPML,
  FUF/SURGE
• Ref http//www.ub.utwente.nl/webdocs/ctit/1/00000
  0c5.pdf

15
Important Issues in Realisation

• Reusability most realisers are
  application-specific, need to be built for each
  new application
• Wide coverage most realisers cover only
  miniscule subsets of a language
• Controlling variation most realisers dont allow
  the same meaning to be expressed in different
  styles, registers, etc.
• Psychological perspective how do humans do it?

16
Reference

• Ehud Reiter Robert Dale (1997) Building Applied
  Natural Language Generation Systems, Sections 6
  and 7 http//ics.mq.edu.au/rdale/publications/pa
  pers/1997/jnle97.pdf

17
Exercises

• The grammar from last weeks exercises is an
  example of realisation as reversed parsing
• It is limited in various ways
• Covers only present tense
• Covers only singular constructions

18
Adding present/past distinction
Present/past, or tense is a property of
events, therefore of verbs and verb
phrases. Present tense (Simple) past
tense like liked dust dusted run ran fi
nd found
19
Adding present/past distinction
Information about tense needs to be added both at
the word and the phrase level, e.g. word(english
,likes,verb,c_like,present). verb_phrase(Language
,E,P,VP,Tense)- verb(Language,Event,Verb,Tens
e), .
20
Adding singular/plural distinction
Singular/plural, or number is a property of
individual words and of larger constituents such
as noun phrases and verb phrases. In English,
determiners and their nouns must both be either
singular or plural, i.e. their number must agree.
The same holds for (subject) noun phrases and
their verb phrases. NP this boy NP these
boys NP these boy NP this boys
21
Adding singular/plural distinction
S the boys drink S the boy drinks S the
boy drink S the boys drinks S the boys
drank S the boy drank Information about
number needs to be added at the word level,
e.g. word(english,boy,noun,c_boy,singular).
22
Adding singular/plural distinction
Information about number also needs to be added
at the phrase level, e.g. noun_phrase(Lang,Conce
pt,NounPhrase,Number)- determiner(Language,Dete
rminer,Number), noun(Language,Concept,Noun,Numbe
r), append(Determiner,Noun,NounPhrase). This
time, the phrase and its two constituents
(determiner and noun) have the same variable as
their last arguments this enforces syntactic
agreement.