There and Back Again: Cross-Lingual Transfer Learning for Event Detection

Tommaso Caselli; Ahmet Üstün

There and Back Again: Cross-Lingual Transfer Learning for Event Detection

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

2 Citations (Scopus)

Abstract

In this contribution we in- vestigate the generalisation abilities of a pre-trained multilingual Language Model, namely Multilingual BERT, in different transfer learning scenarios for event de- tection and classification for Italian and English. Our results show that zero-shot models have satisfying, although not opti- mal, performances in both languages (av- erage F1 higher than 60 for event detec- tion vs. average F1 ranging between 40 and 50 for event classification). We also show that adding extra fine-tuning data of the evaluation language is not simply ben- eficial but results in better models when compared to the corresponding non zero- shot transfer ones, achieving highly com- petitive results when compared to state-of-
the-art systems.

Original language	English
Title of host publication	Proceedings of the Sixth Italian Conference on Computational Linguistics
Publisher	CEUR Workshop Proceedings (CEUR-WS.org)
Volume	2481
Publication status	Published - 2019

Access to Document

http://ceur-ws.org/Vol-2481/paper15.pdfLicence: CC BY

Handle.net

Persistent link

Cite this

@inproceedings{2f311bddbc9543d681caf30a398c2fb5,

title = "There and Back Again: Cross-Lingual Transfer Learning for Event Detection",

abstract = "In this contribution we in- vestigate the generalisation abilities of a pre-trained multilingual Language Model, namely Multilingual BERT, in different transfer learning scenarios for event de- tection and classification for Italian and English. Our results show that zero-shot models have satisfying, although not opti- mal, performances in both languages (av- erage F1 higher than 60 for event detec- tion vs. average F1 ranging between 40 and 50 for event classification). We also show that adding extra fine-tuning data of the evaluation language is not simply ben- eficial but results in better models when compared to the corresponding non zero- shot transfer ones, achieving highly com- petitive results when compared to state-of-the-art systems.",

author = "Tommaso Caselli and Ahmet {\"U}st{\"u}n",

year = "2019",

language = "English",

volume = "2481",

booktitle = "Proceedings of the Sixth Italian Conference on Computational Linguistics",

publisher = "CEUR Workshop Proceedings (CEUR-WS.org)",

}

TY - GEN

T1 - There and Back Again

T2 - Cross-Lingual Transfer Learning for Event Detection

AU - Caselli, Tommaso

AU - Üstün, Ahmet

PY - 2019

Y1 - 2019

N2 - In this contribution we in- vestigate the generalisation abilities of a pre-trained multilingual Language Model, namely Multilingual BERT, in different transfer learning scenarios for event de- tection and classification for Italian and English. Our results show that zero-shot models have satisfying, although not opti- mal, performances in both languages (av- erage F1 higher than 60 for event detec- tion vs. average F1 ranging between 40 and 50 for event classification). We also show that adding extra fine-tuning data of the evaluation language is not simply ben- eficial but results in better models when compared to the corresponding non zero- shot transfer ones, achieving highly com- petitive results when compared to state-of-the-art systems.

AB - In this contribution we in- vestigate the generalisation abilities of a pre-trained multilingual Language Model, namely Multilingual BERT, in different transfer learning scenarios for event de- tection and classification for Italian and English. Our results show that zero-shot models have satisfying, although not opti- mal, performances in both languages (av- erage F1 higher than 60 for event detec- tion vs. average F1 ranging between 40 and 50 for event classification). We also show that adding extra fine-tuning data of the evaluation language is not simply ben- eficial but results in better models when compared to the corresponding non zero- shot transfer ones, achieving highly com- petitive results when compared to state-of-the-art systems.

UR - http://ceur-ws.org/

M3 - Conference contribution

VL - 2481

BT - Proceedings of the Sixth Italian Conference on Computational Linguistics

PB - CEUR Workshop Proceedings (CEUR-WS.org)

ER -

There and Back Again: Cross-Lingual Transfer Learning for Event Detection

Abstract

Access to Document

Handle.net

Other files and links

Fingerprint

Cite this