Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Rule-based Openness Transformer 'OpennessRuleScore' #203

Open
wants to merge 2 commits into
base: master
Choose a base branch
from

Conversation

klj64
Copy link

@klj64 klj64 commented Dec 11, 2023

Description

Feature: This PR introduces a new transformer that scores utterances with questions (0 otherwise).

The rule-based method checks to see if there are keywords associated with "open" or "closed" and adds or subtracts a point respectively. The list of keywords associated with questions is finite. This list includes the wh-questions previously explained through Pomerantz. For example, what, why, and how are all considered indicators that a question is "more open." While who, where, when, and which are all considered "more closed."

This takes from both the distinction that Robinson and Rackstraw make, but also Dohrenwend's idea that closed questions consist of the (1) selection question, (2) yes-no question, and (3) identification question.

Another notable category of indicator words are words that invoke opinion such as "wonder," "think," or "you." These tokens come from the human annotation tasks. We also relegate "yeah," "yes," and "no" as indicators of Dohrenwend's yes-no questions and rhetorical questions.

When checking against the utterance, we only score the last question. This is due to the fact that when asked a series of questions, respondents will usually answer the last one posed. This also helps us avoid scoring multiple questions in one utterance.

Motivation and Context

We seek to examine the phenomena of question "openness" and how it might change throughout a conversation.

How Has This Been Tested?

We ran sanity checks through identifying a few utterances in a corpus (CANDOR) and seeing if the score performs as expected one the transformer is applied.

Other information:

N/A

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant