Discourse Resolution

Question 1

Discourse Relation

Answer 1

Implicit relationship between two sentences. This may be ambiguous e.g. is it a narration or an explanation.

Question 2

Factors influencing discourse interpretation

Answer 2

Cue phrases

Punctuation/intonation and text structure.

Real world content.

Tense and aspect.

Question 3

Summarisation

Answer 3

Can remove explanatory phrases in discourse relation and leave coherent structure still.

Question 4

Referent

Answer 4

A real world entity that some piece of text refers to.

Question 5

Referring Expressions

Answer 5

Bits of language used to perform reference by a speaker.

Question 6

Antecedent

Answer 6

The text initially evoking a referent.

Question 7

Anaphora

Answer 7

The phenomenon of referring to an antecedent.

Question 8

Cataphora

Answer 8

Pronoun appearing before their referents are introduced by a proper name or definite description.

Question 9

Pronoun Agreement

Answer 9

Pronouns generally have to agree in number and gender with their antecedents.

Question 10

Reflexives

Answer 10

Relexive pronouns must be co-referential with a preceding argument of the same verb while non-reflexive pronouns cannot be.

Question 11

Pleonastic pronouns

Answer 11

Semantically empty and don’t refer.

Question 12

Recency

Answer 12

More recent antecedents are preferred. Only relative recently referred to entities are accessible.

Question 13

Grammatical role

Answer 13

Subjects > objects > everything else

Question 14

Repeated Mention

Answer 14

Entities that have been more frequently mentioned are preferred.

Question 15

Parallelism

Answer 15

Entities sharing the same role as the pronoun in the same sort of sentence are preferred.

Question 16

Coherence effects

Answer 16

The pronoun resolution may depend on the rhetorical/discourse relation that is inferred.

Question 17

Anaphora resolution algorithms

Answer 17

Formulate pronoun resolution as a classification problem. For each pairing of a pronoun and a candidate antecedent, the classifier has to make a binary decision as to whether the candidate is an actual antecedent, based on the pairing’s features. These are trained and classified using a feature vector.

Question 18

Pronoun resolution evaluation

Answer 18

Possibility:

Require every pronoun is linked an antecedent and just measure accuracy of links found compared to test data.

Problems: pleonastic pronouns and identification of target NPs due to embedded NPs being a particular issue.