Conversation

Once a model takes part in conversations, you need a way to measure what happens in them, from politeness to structure to how people's language changes. This page points at Cornell's ConvoKit and then walks its feature extractors in order: politeness strategies, expected conversational context, hypergraph structure, and linguistic diversity. Each one pairs the research page with its API documentation.

ConvoKit by Cornell — is a toolkit for extracting conversational features and analyzing social phenomena in conversations. It includes several large conversational datasets along with scripts exemplifying the use of the toolkit on these datasets.

The first feature is politeness. Politeness strategies is Cristian Danescu-Niculescu-Mizil's research on a computational approach to politeness with application to social factors, and the toolkit's version is documented in the politenessStrategies (API). It gives a set of lexical and parse-based features correlating with politeness and impoliteness.

Politeness looks at single utterances; the next framework looks at the context an utterance is expected to appear in. Expected Conversational Context Framework is the dissertation page behind it, and the Expected Context Framework (API) is the documentation. It is a framework for characterizing utterances and terms based on their expected conversational context, consisting of model implementations and wrapper pipelines.

From utterances the toolkit moves to the shape of a whole discussion. Hypergraph conversation representation is Cristian Danescu-Niculescu-Mizil's research on characterizing online public discussions through patterns of participant interactions, and Hyperconvo is the (API). It is a method for extracting structural features of conversations through a hypergraph representation.

The last extractor follows speakers rather than conversations. Linguistic diversity in conversations is Cristian Danescu-Niculescu-Mizil's research "Finding Your Voice: The Linguistic Development of Mental Health Counselors", and Speaker Convo Diversity is the (API). It is a method to compute the linguistic diversity of individuals within their own conversations, and between other individuals in a population.