Article Structure
Abstract
We propose a computational framework for identifying linguistic aspects of politeness.
Introduction
Politeness is a central force in communication, arguably as basic as the pressure to be truthful, informative, relevant, and clear (Grice, 1975; Leech, 1983; Brown and Levinson, 1978).
Politeness data
Requests involve an imposition on the addressee, making them a natural domain for studying the interconnections between linguistic aspects of politeness and social variables.
Politeness strategies
As we mentioned earlier, requests impose on the addressee, potentially placing her in social peril if she is unwilling or unable to comply.
Predicting politeness
We now show how our linguistic analysis can be used in a machine learning model for automatically classifying requests according to politeness.
Relation to social factors
We now apply our framework to studying the relationship between politeness and social variables, focussing on social power dynamics.
Related work
Politeness has been a central concern of modern pragmatic theory since its inception (Grice, 1975; Lakoff, 1973; Lakoff, 1977; Leech, 1983; Brown and Levinson, 1978), because it is a source of pragmatic enrichment, social meaning, and cultural variation (Harada, 1976; Matsumoto, 1988;
Conclusion
We construct and release a large collection of politeness-annotated requests and use it to evaluate key aspects of politeness theory.
Topics
in-domain
- Classification results We evaluate the classifiers both in an in-domain setting, with a standard leave-one-out cross validation procedure, and in a cross-domain setting, where we train on one domain and test on the other (Table 4).Page 6, “Predicting politeness”
- For both our development and our test domains, and in both the in-domain and cross-domain settings, the linguistically informed features give 3-4% absolute improvement over the bag of words model.Page 6, “Predicting politeness”
- While the in-domain results are within 3% of human performance, the greater room for improvement in the cross-domain setting motivates further research on linguistic cues of politeness.Page 6, “Predicting politeness”
- Encouraged by the close-to-human performance of our in-domain classifiers, we use them to assign politeness labels to our full dataset and then compare these labels to independent measures of power and status in our data.Page 6, “Relation to social factors”
- In-domain Cross-domain Train Wiki SE Wiki SE Test Wiki SE SE WikiPage 6, “Relation to social factors”
- Table 4: Accuracies of our two classifiers for Wikipedia (Wiki) and Stack Exchange (SE), for in-domain and cross-domain settings.Page 6, “Relation to social factors”
See all papers in Proc. ACL 2013 that mention in-domain.
See all papers in Proc. ACL that mention in-domain.
Back to top.
SVM
- The BOW classifier is an SVM using a unigram feature representation.6 We consider this to be a strong baseline for this newPage 5, “Predicting politeness”
- is an SVM using the linguistic features listed in Table 3 in addition to the unigram features.Page 6, “Predicting politeness”
- For new requests, we use class probability estimates obtained by fitting a logistic regression model to the output of the SVM (Witten and Frank, 2005) as predicted politeness scores (with values between 0 and l; henceforth politeness, by abuse of language).Page 6, “Predicting politeness”
See all papers in Proc. ACL 2013 that mention SVM.
See all papers in Proc. ACL that mention SVM.
Back to top.
