Coherence Relation

1
Coherence Relation Question-Answer System

• Presented by
• Safeya Mamish

2
Plan

• What is a coherence relation
• (examples-conjunctions - problems)
• Coding procedures and comments
• Sample Run
• Question-Answer
• Previous works

3
Coherence Relations

• Coherence relations are the most important
  factors to establish coherent representation when
  processing a text.
• The coherence relation provides a clue for the
  determination of coherence and discourse
  structure, and hence the larger meaning of the
  text.
• It is an application of sub-tagging

4
Conjunctions of Coherence
Because and so Cause-Effect
Although, but , while Violated expectation
If..(then), as long as, while Condition
And, similarly , as well Similarity
But, however, on the other hand Contrast
Then, first, second, before, after, when Temporal Sequence
According to, said, claim, stated that, Attribution
For example, for instance, how Example
Furthermore, also, in addition, note, Elaboration
In general Generalization
5
Examples of Coherence Relations

• 1-Cause-Effect
• There was bad weather at the airport1 and so
  our flight got delayed.2
• 2-Condition
• If the new software works,1 everyone will be
  happy.2
• 3-Violated Expectation
• The weather was nice1 but our flight got
  delayed.2
• 4-Similarity (parallel)
• The flight 201 was delayed1 Flight 501 arrived
  late as well.2
• 5-Contrast
• The flight 201 was delayed1 Flight 501 arrived
  on time.2

6
Examples of Coherence Relations-2

• 6-Example
• There are several mission to Mars1 Pathfinder
  is an important one.2
• 7-Elaboration
• The new version of the software works well1 It
  is version 5.0.2
• 8-Temporal sequence
• First John went to a grocery1 Then he went to
  the YMCA.2
• 9-Same
• The flight 201, 1 coming from Paris,2
  arrived late.2
• 10-Attribution
• John said1 Flight 501 arrived on time.2

7
Coherence Relations

• Problems of the Coherence Relations
• The identification of a specific coherence
  relation from a given set is not a
  straightforward task, even for people, it may be
  fuzzy . Ex
• 1. John bought a raincoat.
• 2. He went shopping yesterday on Queen
  Street and it rained.
• The coherence relation here could be
  elaboration (on the buying), or explanation (of
  when, how, or why), or cause (he bought the
  raincoat because it was raining out).
• Although the importance of the coherence
  relations, it is not automatically coded.

8
Coding Procedures

• Discourse segments
• Insert boundaries at every period
• Not initial one such as Mr. or Dept.
• Boundaries at every semi-colon or colon
• Boundaries at valid commas
• Neglect comma separating verbs or nouns (He
  likes apples, pears, and bananas)
• Boundaries at every quotation marks
• Boundaries at coherence conjunction

9
Comments on step-1

• Apparently it a straightforward and easy part in
  coding.
• The main problem here is when conjunction with
  comma or and
• Easy detected It snowed and rained all the
  day.
• Harder if it needs logic interpretation
• Milk sold to the nation's dairy plants and
  dealers averaged 14.5 for each hundred pounds

10
Coding Procedures-2

• Discourse segments Grouping
• Group contiguous discourse segments that are
• - enclosed by a pair of quotation marks.
• attributed to the same source.
• belong to the same sentence.
• topically centered on the same events.

11
Comments on step-2

• Apparently it is not a straightforward part in
  coding.
• Access to the concept of lexical chaining,
  lexical cohesion, and with indexes and pointers
  that relates words semantically.
• Example
• 1. Mary spent three hours in the garden
  yesterday.
• 2. he was digging potatoes.
• In a way, there will be a pointer from the word
  digging to the word garden, and thus the
  discourses may be grouped together.

12
Coding Procedures-3

• Identify the Coherence relations between
  Discourse segments
• Some Suggestions
• - Use pair of quotation marks as a signal for
  attribution.
• Colon not followed by QM are explanation.
• Valid Commas on same discourses is Same.
• Repeated keywords as elaboration.

13
Comments on step-3

• Hardest part in coding, cause it may be fuzzy.
• Some conjunctions falls in two or more categories
  such then (condition / temporal ) or but
  (contrast, violated expectation)
• Some techniques that may be used
• - Alignment (Probabilistic Translation)
• - Chain Rule decomposition
• - Lexical representation
• - Synthetic representation
• - Rule-Based and Knowledge framework

14
Sample Run

• The main phrase was
• Mr. Han bought tomatoes, apples and bananas,
  then he went home. The tomatoes were really good
• ..
• First step Dividing according to appropriate
  period
• Now discourse 1
• Mr. Han bought tomatoes, apples and bananas,
  then he went home.
• Now discourse 2
• The tomatoes were really good.
• Searching for comma
• Mr. Han bought tomatoes, apples and bananas 1 1
• 1 2 then he went home.
• 2 1 The tomatoes were really good.
• Coherence found in 1
• Temporal sequence Coherence between discourses
  1 1 and 1 2
• Elaboration Coherence between Discourse 2 and
  1 2

15
Question-Answer

• The goal of question-answer system is to find
  answer in the text. Any question-answering is
  based on
• Answer type selection
• Answer entity annotation
• Information Retrieval
• Answer selection

16
Coherence relations Question-Answer

• Classifying questions types (How, When, Why)
• Answer type selection Finding coherence
  conjunctions /relations that can meet the answer
  
• When (temporal ) Who (elaboration) Why
  (cause)
• Answer entity annotation sub tagging the
  coherence relations, and then matched with the
  type of questions asked
• Information Retrieval matching sentences based
  on coherence relations, narrow the different
  possibilities of answers
• Answer selection Using heuristics

17
Some Previous Works

• DiMLex A lexicon of discourse markers for text
  generation and understanding only marks segment
  where there is conjunction of coherence.
• COPARS COHERENCE PARSING Parse texts into
• Sentences according to coherences, but does not
• reveal the hidden ones.
• TempTag Tag the temporal coherences to provide
• answers to (When) questions.

18
References

• Periodicals
• 1.Alistair Knott, Ted Sanders, Jon Oberlander,
  (2001), Levels of representation in discourse
  relations, Cognitive Linguistics vol. 12 (3), pp
  197-209
• 2. Degand, L.(1998), On Classifying Connectives
  and Coherence Relations. In Coling /ACL Workshop
  on Discourse Relations and Discourse Markers.
  Montreal, Canada, pp. 29-35
• 3. Jane Morris, Graeme Hirstt, (1991) Lexical
  Cohesion Computed by Thesaural Relations as an
  Indicator of the Structure of Text, Computational
  Linguistics, Vol. 17,(1),
• 4. Max Louwerse, (2001) , An analytic and
  cognitive parameterization of coherence
  relations, Cognitive Linguistics vol. 12 (3), pp,
  291-315
• 5. Stokes, N., Carthy, J., Smeaton, A.F. (2002) .
  Segmenting Broadcast News Streams Using Lexical
  Chains. Proceedings of 1st Starting AI
  Researchers Symposium (STAIRS 2002), vol. 1,
  pp.145-154.

19
References (2)

• 6. Ted J. M. Sanders, Leo G. M. Noordman, The
  Role of Coherence Relations and Their Linguistic
  Markers in Text Processing , Discourse Processes,
  Vol. 29(1), 3760
• 7. Van Dijk, Teun A., 1980. An Extended Theory of
  Speech Acts Appropriateness Conditions for
  Subordinate Acts in Sequences, Journal of
  Pragmatics 4, pp. 233-252.
• 8. Wolf, Florian Gibson , Edward (2003) The
  descriptive inadequacy of trees for representing
  discourse coherence, technical report,
  Massachusetts Institute of Technology, Department
  of Brain and Cognitive Sciences, Cambridge, MA,
  USA.
• 9. Wolf, Florian Gibson , Edward (2005)
  Representing discourse coherence, Computer
  Linguistics, Vol(31)-2, June 2005, pp-249-255
• Books
• 10- Sholom M.Weiss, Nitin Indurkhya, Tong Zhang,
  Fred Damerau, Text mining, Springer Science
  Business Media, 2005.