Contrastively focused pronouns

Data used in the Interspeech 2022 paper "BERT, can HE predict contrastive focus? Predicting and controlling prominence in neural TTS using a language model"

This is a corpus of literary texts that have been annotated with prominence and boundary features using the Wavelet Prosody Toolkit (https://github.com/asuni/wavelet_prosody_toolkit). Each text is read by three separate speakers. A subcorpus of contrastively focused pronouns is also provided.

Train and Test sets for prominence prediction task.

-Lines starting with '' identify the utterances. Here you will find book/speaker/chapter/chapter-utterance# information.

-The columns of the remaining lines:

word  /  quantized CWT prominence features / quantized CWT boundary features  /   Raw CWT prominence features  /   Raw CWT boundary features

Majority, Minority and nonContrastivePronouns are dictionaries containing the chapter-utterance# (in Test.txt) and sentence index of the pronouns used for evaluation in the Interspeech paper.

Majority - at least two out of three speakers use contrastive focus.

Minority - only one speaker used contrastive focus.

nonContrastivePronouns - none of the speakers used contrastive focus.