Rhetoric NLP Tasks

I recently have been annotating speeches following up on the edx a course on Rhetoric.

The following NLP task emerge:

Rhetorical Tasks & a Loss function

The Goal

• Extract and specify identify the goal of the speech as a single sentence — using abstractive extraction.
• The goal is important as it set the context for the whole speech, and the subsequent tasks use this context as part of their definition.

Map the structure

Structure is important

• It helps the speaker engage the audience.
• It helps to lay out the arguments in a sensible fashion.
• by building interest and expectation in what would otherwise be torturous experience.
• The order of the sections were often modified.

Classical rhetoric uses the following top level parts of the oration.

a classifier for each paragraph’s structural element.

Table 1: main section used in classical oratory

name	purpose	main mode
exodium	introduction of the speaker and topic	ethos
narratio	statement of facts,	logos
propositio	a summary of what one is about to speak on e.g. lay charge or accusation
partito	division	logos
confirmatio	proof	logos
refutio	refutation	logos
peroratio	conclusion	pathos

The speakers of antiquity went further by employing mnemonic devices by associate each section of the oration with the layout of the venue.

The task here is to maps/align the text to each established sections.

Cicero’s speech structure appears to be predicated on a legal oration in defense of a client.

As our oration or essay diverges from this type of medium we will need to adapt our structure.

Cicero also maps each section to an recommended mode of persuasion.

However, most speakers tend to have sequences of small cluster with all three. (e.g. ethos, logos pathos).

Thus I expect that classiing the sections would be deffered until additional features are identified.

2. for each phrase : classify the main mode of persuasion rough: logos, ethos, pathos fine: telos, mythos, nomos, okios,
3. for each sentence bin classifier for new bliss or problem.
5. Cicero wrote about rhetoric in service of leagal defence - his forte. Political speeches can take other forms.

• Is there a ontology for speeches ?
• How important is understanding structure for writing good speeches.

A Generative approach

Some of the hallmarks of good writing are:

1. Every part serves a purpose - given a goal we consider this goal when ever we write.
2. It is dense with patterns. By which I mean each part is often serving more than one purpose simultaneously
3. There us a unity of parts. In this case there is a structure
4. There are synergies - the whole is greater than the sum of its parts.

A generative approach for getting this:

1. a plan - using frames or otherwise create a structure for the speech/discourse.

• break these down into more detailed instructions perhaps using a language model - resulting in a tree of retorical specs
• use these specs as prompts to generate the speech.

2. provide the model with a loss function - to measure how effective the generator is.

• the loss function would be built on the above