Photo by — @deeezyfree

Five Types of Structure in Computational Models of Discourse

Discourse in Computational Linguistics and Artificial Intelligence

This article is a short extract of an aspect of the text Discourse in Computational Linguistics and Artificial Intelligence written by Johanna D. Moore and Peter Wiemer-Hastings. I will try to sum up the five.

According to the article researchers in computational linguistics have argued that coherent discourse has structure.

Recognising this structure is a crucial component of comprehending the discourse.

They argue that there is consensus among researchers in computational linguistics that at least three types of structure are needed in computational models of discourse processing.

Those three are:

  1. Intentional structure.
  2. Informational structure.
  3. Attentional structure.

They added two more, that I will sum up towards the end. Those two are (4) information structure and (5) rhetorical structure.

1. Intentional structure

The roles that utterances play in the speaker’s communicate plan. The speaker plans to achieve a desired effect on the hearer’s mental state or conversational record.

Intentions can encode what the speaker wanted to accomplish with a certain discourse.

Without the intentional structure it is hard or impossible to clarify an utterance.

Intentions are important in nominal expressions and discourse cues (such as ‘because’ and ‘thus’) or scalar terms (like ‘difficult’ or ‘easy’).

2. Informational structure

Semantic relationships between information in successive utterances.

→ Causal relations are a typical example of informational structure.

Researchers have discovered inherent structure of the subject matter being communicated.

This can be of a certain domain depending on relationships between words.

There are spatial, familial or causal relations between objects or events being described.

3. Attentional structure

Humans “focus” or “center” their attention on a small set of entities and attention shifts to new entities in predictable ways.

Systems of understanding in natural language must track attentional shifts.

This must be learned to resolve anaphoric expressions.

“In linguistics, anaphora is the use of an expression whose interpretation depends upon another expression in context.”

“a. Susan dropped the plate. It shattered loudly. — the pronoun it is an anaphor; it points to the left toward its antecedent the plate.

b. The music stopped, and that upset everyone. — The demonstrative pronoun that is an anaphor; it points to the left toward its antecedent The music stopped.”

It is important as well to understand ellipsis: the omission from speech or writing a word or words that are superfluous or able to be understood from contextual clues.

One can construct individual responses in order to influence choices on what to say next.

So, those are the three with consensus, and they added a further two.

4. Information structure

There are two dimensions to information structure:

  1. “The contrast a speaker makes between the part of an utterance that connects it to the rest of the discourse (the theme), and the part of an utterance that contributes new information on that theme (the rheme).
  2. What the speaker takes to be in contrast with things a hearer is or can be attending to.”

This focus within the compositional semantics forms a part of formal grammar.

Rheme (or focus): the part of a clause that gives information about the theme → what is being talked about.

In linguistics, what is being said about a topic, is theme–rheme.

Topic and subject are also distinct concepts from agent (or actor) — the “doer”, which is defined by semantics.

5. Rhetorical structure

This notion within computational linguistics can be used to explain a wide range of discourse phenomena.

There has been attempts to explain the inferences that arise when a particular relation holds between two discourse entities.

The relation may not be explicitly signalled in the text.

“Researchers in generation have shown that it is crucial that a system recognise the additional inferences that will conveyed by the sequence of clauses they generate, because these additional inferences may be the source of problems if the user does not understand or accept the system’s utterance.”

Many text generation systems have used these patterns to construct coherent monologic texts to achieve a variety of discourse purposes.


Moore, J. D., & Wiemer-Hastings, P. (2003). Discourse in computational linguistics and artificial intelligence. Handbook of discourse processes, 439–486.

This is #500daysofAI and you are reading article 409. I am writing one new article about or related to artificial intelligence every day for 500 days.



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Alex Moltzau

Alex Moltzau

AI Policy and Ethics at Student at University of Copenhagen MSc in Social Data Science. All views are my own.