Optimizing Referential Coherence in Text Generation Rodger Kibble and Richard Power 2004

1
Optimizing Referential Coherence in Text
GenerationRodger Kibble and Richard Power (2004)

• Irina Borisova

2
Outline

• Review
• Theoretical base
• - rhetorical structure theory
• - centering theory
• Generation procedure
• - enumerating textual structures
• - enumerating centers
• - evaluation
• - Conclusions

3
The articleReview

• Objectives
• to analyze the influence of referential
  continuity on the global coherence
• to describe a method of applying centering theory
  to the issues of text planning and
  pronominalization in order to improve the quality
  of generated texts
• Novelty referential coherence is treated as a
  planning problem.
• Benefit can be used as a research tool for the
  evaluation of the centering theory

4
Restrictions

• Assumptions
• one-to-one correspondence between predicates and
  verbs
• - pronominalization is delayed until grammatical
  relations and word order have been determined
• ICONOCLAST implemented document generation
  system
• constraint satisfaction
• user-determined rhetorical structures are
  transformed into text structure

5
Referential coherence

• a. Elixir is a white cream.
• It is used in the treatment of cold sores.
  
• It contains aliprosan.
• Aliprosan relieves viral skin disorders.
• b. Elixir contains aliprosan.
• Viral skin disorders are relieved by
  aliprosan.
• Elixir is used in the treatment of cold
  sores.
• It is a white cream.

6
Rhetorical Structure Theory (1)

• Rhetorical Structure Theory (Mann and Thompson,
  1987)
• 23 rhetorical structures
• nucleus the more central unit from the writers
  purpose is interpretable independently
• satellite less central is interpretable with
  respect to nucleus
• - rhetorical structure is represented as a set of
  constraints on the nucleus and satellite

7
Rhetorical Structure Theory (2)

• Contrast
• This is a multinuclear relation, in which two or
  more nuclei contrast along some important
  dimension.
• The priest was in a very bad temper, but the lama
  was quite happy.

8
Rhetorical Structure Theory (3)
9
Centering Theory (1) (Grosz and Sidner, 1986,
Grosz et al, 1995)

• Discourse structure
• linguistic structure (discourse segments)
• intentional structure (intentions and relations
  between discourse segments)
• - attentional state (attention at any given point
  of discourse)

10
Centering theory (2)Basic Definitions

• Centers the entities that link the utterance to
  other utterance within the discourse segment.
• discourse construct
• semantic object
• Each utterance U has a set of forward-looking
  centers Cf, each utterance U other than initial
  has a single backward-looking center Cb. The
  backward-looking center of utterance Un1
  connects with one of the forward-looking centers
  of utterance Un.
• Kibble and Power preferred center Cp

11
Centering parameters (1)Ranking and Utterance
Realization

• Ranking subject gt direct object gt indirect
  object gt others.
• Assumptions
• syntactic realization works to determine Cp from
  other components at the same level, so subject gt
  others
• the system already knows, which entities can
  occur in the subject position
• Utterances
• Kameyama (1998) model local focus is influenced
  by clauses rather than by sentences
• Suri, McCoy, DeCristoforo the subject of the
  main clause in the complex sentence is likely to
  be pronominalized in an immediately following
  sentence

12
Centering parameters (2)Ranking and Utterance
Realization

• Utterances
• Kameyama (1998) model local focus is influenced
  by clauses rather than by sentences
• Suri, McCoy, DeCristoforo the subject of the
  main clause in the complex sentence is likely to
  be pronominalized in an immediately following
  sentence
• Veins theory (adopted by Kibble and Power)

13
Centering parameters (3)Transitions and
constraints

• Traditional transition relations
• - Center continuation Cb(Un1) Cb(Un) the
  most highly ranked entity in Cf(Un1)
• - Center retaining Cb(Un1) Cb(Un) not the
  most highly ranked entity in Cf(Un1)
• - Center shifting Cb(Un1) ? Cb(Un)

14
Centering parameters (4)Transitions and
constraints

• cohesion Cb(Un-1) Cb(Un)
• salience Cp(Un ) Cb(Un )
• cheapness Cp(Un-1 ) Cb(Un )
• continuity Cfs(Un-1 ) ? Cfs(Un )?? Ø

15
Centering parameters (5) Preferences
Pronominalization and Sequences

• Rule 1 If any element of Cf(Un) is realized by a
  pronoun in Un1, then the Cb(Un1) must be
  realized by a pronoun also.
• Kibble and Power The user can choose between
  pronominalizing Cb or using an algorithm based on
  parallelism of grammatical roles.
• Rule 2 Sequences of continuation are preferred
  over sequences of retaining and sequences of
  retaining are preferred over sequences of
  shifting.
• Kibble and Power in order to minimize violations
  of the constraints continuity, cohesion, salience
  and cheapness, violations are computed and summed
  for each candidate to choose one with fewest.

16
Generation Architecture(Reiter 1994)

• Text planning/content determination deciding the
  content of a message and organizing the component
  propositions into a text structure (typically a
  tree)
• Sentence planning aggregating propositions into
  clausal units and choosing lexical items
  corresponding to concepts in the knowledge base
  this is the level at which the order of arguments
  and choice of referring expressions will be
  determined
• Linguistic realization surface details such as
  agreement and orthography

17
Generation Architectureand Centering Theory

• Incremental sentence-by-sentence generation
• A pipelined system
• Identifying Cbs and Cps is an optimization
  problem

18
Generation ProcedureOutline

• Enumerate all text structures that are acceptable
  realizations for the rhetorical structure.
• For each text structure, enumerate all
  permissible choices for the Cb and Cp of each
  proposition.
• Evaluate the solutions, taking account of
  referential coherence among other considerations,
  and choose the best.

19
Text PlannerICONOCLAST
20
Generation ArchitectureEnumerating Text
Structures (1)

• a text structure as an ordered tree where each
  node has a feature called TEXT-LEVEL.
• TEXT-LEVEL values
• 0 text phrase
• 1 text clause
• 2 text sentence
• 3 paragraph
• 4 section

21
Generation Architecture Enumerating Text
Structures (1)

• Well-formed text structure should follow the
  hierarchy and must realize the rhetorical
  structure correctly
• The nodes of the text structure should express
  all the elementary propositions in the rhetorical
  structure (also discourse connectives
  optionally).
• The text structure must follow rules of syntax
  when combines propositions and discourse
  connectives within a text clause.
• The text structure must be structurally
  compatible with the rhetorical structure (same
  grouping of the elementary propositions)

22
Generation ArchitectureEnumerating centering
assignments

• Determine the antecedent Un-1 of each proposition
  Un.
• List all potential Cbs and Cps of each
  proposition.
• Compute all combinations from ?Cb and ?Cp that
  respect the fundamental centering constraint that
  Cb(Un) should be the most salient candidate in
  Un-1.

23
Generation ArchitectureEvaluation

• - Compute estimated weight for each type of
  violation (1 to 3)
• - Compute the total cost for the solution
• - Find the solution with the lowest cost

24
Conclusions

• - For systems which generate a single text plan,
  centering theory can determine the most coherent
  ordering of arguments within clauses.
• - For systems which generate multiple text
  plans, centering theory can be used to evaluate
  the different plans as well as to determine the
  optimal realization of any particular plan.