Discourse Coherence

Ever read something that just didn't "flow"? That's likely a problem with discourse coherence. It's what makes a text more than just a random collection of sentences. It's about how sentences and ideas connect to create a unified and meaningful whole. Let's dive into what makes a discourse coherent.

What is Discourse Coherence?

Discourse coherence refers to the way a text's components (sentences, paragraphs, etc.) logically and semantically connect with each other. A coherent discourse is easy to follow and understand because it presents information in an organized and related manner. There are several factors that make a discourse coherent, which include: coherence relations between the sentences, entity-based coherence, and topical coherence.

Coherence Relations

Coherence relations define the logical and semantic connections between different parts of a text. These relations help readers understand how different ideas relate to each other. There are many types of coherence relations, with two prominent frameworks being Rhetorical Structure Theory (RST) and the Penn Discourse TreeBank (PDTB).

Rhetorical Structure Theory (RST)

RST focuses on hierarchical relationships between text spans. It identifies a nucleus (the main part) and a satellite (supporting information). Consider this example:

"Kevin must be here. His car is parked outside."

Here, "Kevin must be here" is the nucleus, and "His car is parked outside" is the satellite, providing evidence. RST relations can be represented graphically with arrows indicating the nucleus-satellite relationship.

Some common RST relations include:

• Attribution: The satellite gives the source of attribution for an instance of reported speech in the nucleus.
  Example: [SAT Analysts estimated] [NUC that sales at U.S. stores declined in the quarter, too]
• List: In this multinuclear relation, a series of nuclei is given, without contrast or explicit comparison:
  Example: [NUC Billy Bones was the mate; ] [NUC Long John, he was quartermaster]
• Evidence: The satellite gives evidence that proves or validates the nucleus.
  Example: Kevin must be here. His car is parked outside.

Penn Discourse TreeBank (PDTB)

PDTB takes a different approach, focusing on discourse connectives—words that signal discourse relations like "because," "although," or "as a result." It annotates these connectives and the text spans they connect (Arg1 and Arg2). For example:

Jewelry displays in department stores were often cluttered and uninspired. And the merchandise was, well, fake. As a result, marketers of faux gems steadily lost space in department stores to more fashionable rivals—cosmetics makers.

Here, "as a result" signals a causal relationship between the first two sentences (Arg1) and the third sentence (Arg2).

PDTB also annotates implicit relations, where no explicit connective is present, but a relation can be inferred. The annotator chooses the word or phrase that could have been the signal (in this case as a result), and then labels its sense.

Example:

In July, the Environmental Protection Agency imposed a gradual ban on virtually all uses of asbestos. (implicit=as a result) By 1997, almost all remaining uses of cancer-causing asbestos will be outlawed.

PDTB includes a tagset with classes, types, and subtypes, such as:

Class	Type	Example
TEMPORAL	SYNCHRONOUS	The parishioners of St. Michael and All Angels stop to chat at the church door, as members here always have. (Implicit while) In the tower, five men and women pull rhythmically on ropes attached to the same five bells that first sounded here in 1614.
CONTINGENCY	REASON	Also unlike Mr. Ruder, Mr. Breeden appears to be in a position to get somewhere with his agenda. (implicit=because) As a former White House aide who worked closely with Congress, he is savvy in the ways of Washington.
COMPARISON	CONTRAST	The U.S. wants the removal of what it perceives as barriers to investment; Japan denies there are real barriers.
EXPANSION	CONJUNCTION	Not only do the actors stand outside their characters and make it clear they are at odds with them, but they often literally stand on their heads.

Entity-Based Coherence

Another aspect of coherence is maintaining focus on specific entities throughout the discourse. Two models that capture entity-based coherence are Centering Theory and the Entity Grid model.

Centering Theory

Centering Theory proposes that at any given point in the discourse one of the entities in the discourse model is salient: it is being “centered” on. As a model of discourse coherence, Centering proposes that discourses in which adjacent sentences CONTINUE to maintain the same salient entity are more coherent than those which SHIFT back and forth between multiple entities.

Centering Theory helps track how entities are introduced and referenced to maintain a clear focus. It uses concepts like:

• Backward-looking center (Cb): The most salient entity in the current utterance that was also mentioned in the previous utterance.
• Forward-looking centers (Cf): A ranked list of entities mentioned in the current utterance that could become the Cb in the next utterance.
• Preferred center (Cp): The highest-ranked entity in the Cf list, indicating what the speaker is most likely to talk about next.

These concepts help determine the intersentential relationships between a pair of utterances Un and Un+1 that depend on the relationship between Cb(Un+1), Cb(Un), and Cp(Un+1)

Centering Theory helps determine transition states. Some possible transitions are:

Continue: The most coherent transition, occurs when Cb(Un+1) = Cb(Un) and Cb(Un+1) = Cp(Un+1)

Retain: Cb(Un+1) = Cb(Un), but Cb(Un+1) ̸= Cp(Un+1)

Smooth-Shift: Cb(Un+1) = Cp(Un+1), but Cb(Un+1) ̸= Cb(Un)

Rough-Shift: Cb(Un+1) ̸= Cb(Un) and Cb(Un+1) ̸= Cp(Un+1)

Example:

a. John went to his favorite music store to buy a piano.

b. He was excited that he could finally buy a piano.

Here, "John" is the backward-looking center, and is also the preferred center, making this a 'Continue' transition and highly coherent.

Entity Grid Model

The entity grid model is an alternative way to capture entity-based coherence by inducing the patterns of entity mentioning that make a discourse more coherent. Instead of having a top-down theory, the entity-grid model using machine learning to induce the patterns of entity mentioning that make a discourse more coherent.

This model uses a grid where rows represent sentences, columns represent discourse entities, and cells indicate the grammatical role of each entity in each sentence (e.g., subject, object, neither, or absent).

Coherence is measured by patterns of local entity transition. For example, Department is a subject in sentence 1, and then not men- tioned in sentence 2; this is the transition [S –]. The transitions are thus sequences {S,O X, –}n which can be extracted as continuous cells from each column.

Topical Coherence

This aspect refers to maintaining a consistent topic or semantic field throughout the discourse. Sentences should relate to the main subject matter and use vocabulary that aligns with the overall theme.

For example, a discussion about climate change should consistently use terms related to environmental science, global warming, and related concepts.

Global Coherence

A discourse must also cohere globally rather than just at the level of pairs of sen- tences. Consider stories, for example. The narrative structure of stories is one of the oldest kinds of global coherence to be studied. Theories such as arguments, and scientific discourse are widely studied computationally.

Argumentation Structure

The first type of global discourse structure is the structure of arguments. Analyzing people’s argumentation computationally is often called argumentation mining. Persuasive essays often rely on a main claim, which is presented and then defended by premises, which support or attack the main claim. Argumentative relations are the relations between the claims and premises.

Example:

“(1) Museums and art galleries provide a better understanding about arts than Internet. (2) In most museums and art galleries, de- tailed descriptions in terms of the background, history and author are provided. (3) Seeing an artwork online is not the same as watching it with our own eyes, as (4) the picture online does not show the texture or three-dimensional structure of the art, which is important to study.”

Thus this example has three argumentative relations: SUPPORT(2,1), SUPPORT(3,1) and SUPPORT(4,3).

The structure of scientific discourse

Scientific papers have a very specific global structure: somewhere in the course of the paper the authors must indicate a scientific goal, develop a method for a solu- tion, provide evidence for the solution, and compare to prior work. Argumentative Zoning is a way of modeling these rhetorical goals, and is informed by the idea that each scientific paper tries to make a knowledge claim about a new piece of knowledge being added to the repository of the field

There are 15 labels related to Argumentative Zoning. A few shortened examples of those labels are:

Category	Description	Example
AIM	Statement of specific research goal, or hypothesis of current paper	“The aim of this process is to examine the role that training plays in the tagging process”
OWN METHOD	New Knowledge claim, own work: methods	“In order for it to be useful for our purposes, the following extensions must be made:”
OWN RESULTS	Measurable/objective outcome of own work	“All the curves have a generally upward trend but always lie far below backoff (51% error rate)”
USE	Other work is used in own work	“We use the framework for the allocation and transfer of control of Whittaker....”

Conclusion

Discourse coherence is crucial for effective communication. By understanding and applying the principles of coherence relations, entity-based coherence, and topical coherence, you can create texts that are clear, logical, and easy for your audience to understand.