Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations

Aarne Talman; Antti Suni; Hande Celikkanat; Sofoklis Kakouros; Jörg Tiedemann; Martti Vainio

Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations

Aarne Talman, Antti Suni, Hande Celikkanat, Sofoklis Kakouros, Jörg Tiedemann, Martti Vainio

Forskningsoutput: Kapitel i bok/rapport/konferenshandling › Konferensbidrag › Vetenskaplig › Peer review

Sammanfattning

In this paper we introduce a new natural language processing dataset and benchmark for predicting prosodic prominence from written text. To our knowledge this will be the largest publicly available dataset with prosodic labels. We describe the dataset construction and the resulting benchmark dataset in detail and train a number of different models ranging from feature-based classifiers to neural network systems for the prediction of discretized prosodic prominence. We show that pre-trained contextualized word representations from BERT outperform the other models even with less than 10% of the training data. Finally we discuss the dataset in light of the results and point to future research and plans for further improving both the dataset and methods of predicting prosodic prominence from text. The dataset and the code for the models are publicly available.

Originalspråk	engelska
Titel på värdpublikation	22nd Nordic Conference on Computational Linguistics (NoDaLiDa) : Proceedings of the Conference
Redaktörer	Mareike Hartmann, Barbara Plank
Antal sidor	10
Utgivningsort	Linköping
Förlag	Linköping University Electronic Press
Utgivningsdatum	30 sep. 2019
Sidor	281–290
ISBN (elektroniskt)	978-91-7929-995-8
Status	Publicerad - 30 sep. 2019
MoE-publikationstyp	A4 Artikel i en konferenspublikation
Evenemang	Nodalida - Nordic Conference on Computational Linguistics - Turku, Finland Varaktighet: 30 sep. 2019 → 2 okt. 2019 Konferensnummer: 22 https://nodalida2019.org/

Publikationsserier

Namn	Linköping Electronic Conference Proceedings
Förlag	Linköping University Electronic Press
Nummer	167
ISSN (tryckt)	1650-3686
ISSN (elektroniskt)	1650-3740

Namn	NEALT Proceedings Series
Nummer	42

Vetenskapsgrenar

113 Data- och informationsvetenskap
6121 Språkvetenskaper

Åtkomst av dokument

W19-6129Slutlig publicerad version, 572 KBLicens: CC BY

Citera det här

Talman, A., Suni, A., Celikkanat, H., Kakouros, S., Tiedemann, J., & Vainio, M. (2019). Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations. I M. Hartmann, & B. Plank (Red.), 22nd Nordic Conference on Computational Linguistics (NoDaLiDa): Proceedings of the Conference (s. 281–290). (Linköping Electronic Conference Proceedings; Nr. 167), (NEALT Proceedings Series; Nr. 42). Linköping University Electronic Press.

Talman, Aarne ; Suni, Antti ; Celikkanat, Hande et al. / Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations. 22nd Nordic Conference on Computational Linguistics (NoDaLiDa): Proceedings of the Conference. redaktör / Mareike Hartmann ; Barbara Plank. Linköping : Linköping University Electronic Press, 2019. s. 281–290 (Linköping Electronic Conference Proceedings; 167). (NEALT Proceedings Series; 42).

@inproceedings{e7f78b80371d4925a8b666b34de47073,

title = "Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations",

abstract = "In this paper we introduce a new natural language processing dataset and benchmark for predicting prosodic prominence from written text. To our knowledge this will be the largest publicly available dataset with prosodic labels. We describe the dataset construction and the resulting benchmark dataset in detail and train a number of different models ranging from feature-based classifiers to neural network systems for the prediction of discretized prosodic prominence. We show that pre-trained contextualized word representations from BERT outperform the other models even with less than 10% of the training data. Finally we discuss the dataset in light of the results and point to future research and plans for further improving both the dataset and methods of predicting prosodic prominence from text. The dataset and the code for the models are publicly available.",

keywords = "113 Computer and information sciences, Natural language processing, 6121 Languages",

author = "Aarne Talman and Antti Suni and Hande Celikkanat and Sofoklis Kakouros and J{\"o}rg Tiedemann and Martti Vainio",

year = "2019",

month = sep,

day = "30",

language = "English",

series = "Link{\"o}ping Electronic Conference Proceedings",

publisher = "Link{\"o}ping University Electronic Press",

number = "167",

pages = "281–290",

editor = "Mareike Hartmann and Plank, {Barbara }",

booktitle = "22nd Nordic Conference on Computational Linguistics (NoDaLiDa)",

address = "Sweden",

note = "Nordic Conference on Computational Linguistics, NoDaLiDa ; Conference date: 30-09-2019 Through 02-10-2019",

url = "https://nodalida2019.org/",

}

Talman, A , Suni, A , Celikkanat, H , Kakouros, S , Tiedemann, J & Vainio, M 2019, Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations. i M Hartmann & B Plank (red), 22nd Nordic Conference on Computational Linguistics (NoDaLiDa): Proceedings of the Conference. Linköping Electronic Conference Proceedings, nr. 167, NEALT Proceedings Series, nr. 42, Linköping University Electronic Press, Linköping, s. 281–290, Nodalida - Nordic Conference on Computational Linguistics, Turku, Finland, 30/09/2019.

Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations. / Talman, Aarne ; Suni, Antti ; Celikkanat, Hande et al.
22nd Nordic Conference on Computational Linguistics (NoDaLiDa): Proceedings of the Conference. red. / Mareike Hartmann; Barbara Plank. Linköping: Linköping University Electronic Press, 2019. s. 281–290 (Linköping Electronic Conference Proceedings; Nr. 167), (NEALT Proceedings Series; Nr. 42).

Forskningsoutput: Kapitel i bok/rapport/konferenshandling › Konferensbidrag › Vetenskaplig › Peer review

TY - GEN

T1 - Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations

AU - Talman, Aarne

AU - Suni, Antti

AU - Celikkanat, Hande

AU - Kakouros, Sofoklis

AU - Tiedemann, Jörg

AU - Vainio, Martti

N1 - Conference code: 22

PY - 2019/9/30

Y1 - 2019/9/30

N2 - In this paper we introduce a new natural language processing dataset and benchmark for predicting prosodic prominence from written text. To our knowledge this will be the largest publicly available dataset with prosodic labels. We describe the dataset construction and the resulting benchmark dataset in detail and train a number of different models ranging from feature-based classifiers to neural network systems for the prediction of discretized prosodic prominence. We show that pre-trained contextualized word representations from BERT outperform the other models even with less than 10% of the training data. Finally we discuss the dataset in light of the results and point to future research and plans for further improving both the dataset and methods of predicting prosodic prominence from text. The dataset and the code for the models are publicly available.

AB - In this paper we introduce a new natural language processing dataset and benchmark for predicting prosodic prominence from written text. To our knowledge this will be the largest publicly available dataset with prosodic labels. We describe the dataset construction and the resulting benchmark dataset in detail and train a number of different models ranging from feature-based classifiers to neural network systems for the prediction of discretized prosodic prominence. We show that pre-trained contextualized word representations from BERT outperform the other models even with less than 10% of the training data. Finally we discuss the dataset in light of the results and point to future research and plans for further improving both the dataset and methods of predicting prosodic prominence from text. The dataset and the code for the models are publicly available.

KW - 113 Computer and information sciences

KW - Natural language processing

KW - 6121 Languages

M3 - Conference contribution

T3 - Linköping Electronic Conference Proceedings

SP - 281

EP - 290

BT - 22nd Nordic Conference on Computational Linguistics (NoDaLiDa)

A2 - Hartmann, Mareike

A2 - Plank, Barbara

PB - Linköping University Electronic Press

CY - Linköping

T2 - Nordic Conference on Computational Linguistics

Y2 - 30 September 2019 through 2 October 2019

ER -

Talman A , Suni A , Celikkanat H , Kakouros S , Tiedemann J , Vainio M. Predicting Prosodic Prominence from Text with Pre-trained Contextualized Word Representations. I Hartmann M, Plank B, redaktörer, 22nd Nordic Conference on Computational Linguistics (NoDaLiDa): Proceedings of the Conference. Linköping: Linköping University Electronic Press. 2019. s. 281–290. (Linköping Electronic Conference Proceedings; 167). (NEALT Proceedings Series; 42).