Every utterance in the dataset was processed with Stanford CoreNLP 3.4.1 ( 28) to generate sentence and word segmentation, part-of-speech tags, and dependency parses used for feature extraction and analysis. After transcription, transcripts were manually cleaned up, heuristically fixing transcriber diarization errors, and correcting typographical errors involving utterance timing so that all transcripts were automatically readable. We used the diarization to automatically remove all officer speech to the dispatcher or to other officers, leaving only speech from the officer directed toward the community member. Transcribers also “diarized” the text (labeling who was speaking at each time point). Extensive measures were taken to preserve privacy data were kept on a central server, and transcribers (as well as all researchers) underwent background checks with the Oakland Police Department. The video for each traffic stop was transcribed into text by professional transcribers, who transcribed while listening to audio and watching the video. We find strong evidence that utterances spoken to white community members are consistently more respectful, even after controlling for contextual factors such as the severity of the offense or the outcome of the stop. In study 3, we apply these models to all vehicle stop interactions between officers of the Oakland Police Department and black/white community members during the month of April 2014. We discuss the linguistic features that contribute to each model, finding that particular forms of politeness are implicated in perceptions of respect. In study 2, we build statistical models capable of predicting aspects of respect based on linguistic features derived from theories of politeness, power, and social distance. Even though they were not told the race of the stopped driver, participants judged officer language directed toward black motorists to be less respectful than language directed toward whites. With a high degree of agreement, participants inferred these dimensions from officer language. In study 1, human participants rated officer utterances on several overlapping dimensions of respect.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |