Morfessor-enriched features and multilingual training for canonical morphological segmentation

Aku Rouhe; Stig-Arne Grönroos; Sami Virpioja; Mathias Creutz; Mikko Kurimo

doi:10.18653/v1/2022.sigmorphon-1.16

Morfessor-enriched features and multilingual training for canonical morphological segmentation

Aku Rouhe, Stig-Arne Grönroos, Sami Virpioja, Mathias Creutz, Mikko Kurimo

Department of Information and Communications Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

67 Downloads (Pure)

Abstract

In our submission to the SIGMORPHON 2022 Shared Task on Morpheme Segmentation, we study whether an unsupervised morphological segmentation method, Morfessor, can help in a supervised setting. Previous research has shown the effectiveness of the approach in semisupervised settings with small amounts of labeled data. The current tasks vary in data size: the amount of word-level annotated training data is much larger, but the amount of sentencelevel annotated training data remains small. Our approach is to pre-segment the input data for a neural sequence-to-sequence model with the unsupervised method. As the unsupervised method can be trained with raw text data, we use Wikipedia to increase the amount of training data. In addition, we train multilingual models for the sentence-level task. The results for the Morfessor-enriched features are mixed, showing benefit for all three sentencelevel tasks but only some of the word-level tasks. The multilingual training yields considerable improvements over the monolingual sentence-level models, but it negates the effect of the enriched features

Original language	English
Title of host publication	Proceedings of the 19th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology
Publisher	Association for Computational Linguistics
Pages	144-151
Number of pages	8
ISBN (Print)	978-1-955917-82-7
DOIs	https://doi.org/10.18653/v1/2022.sigmorphon-1.16
Publication status	Published - 2022
MoE publication type	A4 Conference publication
Event	Meeting of the Special Interest Group on Computational Morphology and Phonology - Seattle, United States Duration: 14 Jul 2022 → 14 Jul 2022 Conference number: 19

Conference

Conference	Meeting of the Special Interest Group on Computational Morphology and Phonology
Abbreviated title	SIGMORPHON
Country/Territory	United States
City	Seattle
Period	14/07/2022 → 14/07/2022

Access to Document

10.18653/v1/2022.sigmorphon-1.16Licence: CC BY

2022.sigmorphon-1.16Final published version, 279 KBLicence: CC BY

USSEE: Understanding Speech and Scene with Ears and Eyes
Kurimo, M. (Principal investigator), Virkkunen, A. (Project Member) & Grósz, T. (Project Member)
01/01/2022 → 31/12/2024
Project: Academy of Finland: Other research funding

Cite this

Rouhe, A., Grönroos, S.-A., Virpioja, S., Creutz, M., & Kurimo, M. (2022). Morfessor-enriched features and multilingual training for canonical morphological segmentation. In Proceedings of the 19th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology (pp. 144-151). Association for Computational Linguistics. https://doi.org/10.18653/v1/2022.sigmorphon-1.16

@inproceedings{8d065664ddea4144bb6eef678119ab2a,

title = "Morfessor-enriched features and multilingual training for canonical morphological segmentation",

abstract = "In our submission to the SIGMORPHON 2022 Shared Task on Morpheme Segmentation, we study whether an unsupervised morphological segmentation method, Morfessor, can help in a supervised setting. Previous research has shown the effectiveness of the approach in semisupervised settings with small amounts of labeled data. The current tasks vary in data size: the amount of word-level annotated training data is much larger, but the amount of sentencelevel annotated training data remains small. Our approach is to pre-segment the input data for a neural sequence-to-sequence model with the unsupervised method. As the unsupervised method can be trained with raw text data, we use Wikipedia to increase the amount of training data. In addition, we train multilingual models for the sentence-level task. The results for the Morfessor-enriched features are mixed, showing benefit for all three sentencelevel tasks but only some of the word-level tasks. The multilingual training yields considerable improvements over the monolingual sentence-level models, but it negates the effect of the enriched features",

author = "Aku Rouhe and Stig-Arne Gr{\"o}nroos and Sami Virpioja and Mathias Creutz and Mikko Kurimo",

note = "| openaire: EC/H2020/771113/EU//FoTran ; Meeting of the Special Interest Group on Computational Morphology and Phonology, SIGMORPHON ; Conference date: 14-07-2022 Through 14-07-2022",

year = "2022",

doi = "10.18653/v1/2022.sigmorphon-1.16",

language = "English",

isbn = "978-1-955917-82-7",

pages = "144--151",

booktitle = "Proceedings of the 19th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology",

publisher = "Association for Computational Linguistics",

address = "United States",

}

Rouhe, A, Grönroos, S-A, Virpioja, S, Creutz, M & Kurimo, M 2022, Morfessor-enriched features and multilingual training for canonical morphological segmentation. in Proceedings of the 19th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology. Association for Computational Linguistics, pp. 144-151, Meeting of the Special Interest Group on Computational Morphology and Phonology, Seattle, Washington, United States, 14/07/2022. https://doi.org/10.18653/v1/2022.sigmorphon-1.16

Morfessor-enriched features and multilingual training for canonical morphological segmentation. / Rouhe, Aku; Grönroos, Stig-Arne; Virpioja, Sami et al.
Proceedings of the 19th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology. Association for Computational Linguistics, 2022. p. 144-151.

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

TY - GEN

T1 - Morfessor-enriched features and multilingual training for canonical morphological segmentation

AU - Rouhe, Aku

AU - Grönroos, Stig-Arne

AU - Virpioja, Sami

AU - Creutz, Mathias

AU - Kurimo, Mikko

N1 - Conference code: 19

PY - 2022

Y1 - 2022

N2 - In our submission to the SIGMORPHON 2022 Shared Task on Morpheme Segmentation, we study whether an unsupervised morphological segmentation method, Morfessor, can help in a supervised setting. Previous research has shown the effectiveness of the approach in semisupervised settings with small amounts of labeled data. The current tasks vary in data size: the amount of word-level annotated training data is much larger, but the amount of sentencelevel annotated training data remains small. Our approach is to pre-segment the input data for a neural sequence-to-sequence model with the unsupervised method. As the unsupervised method can be trained with raw text data, we use Wikipedia to increase the amount of training data. In addition, we train multilingual models for the sentence-level task. The results for the Morfessor-enriched features are mixed, showing benefit for all three sentencelevel tasks but only some of the word-level tasks. The multilingual training yields considerable improvements over the monolingual sentence-level models, but it negates the effect of the enriched features

AB - In our submission to the SIGMORPHON 2022 Shared Task on Morpheme Segmentation, we study whether an unsupervised morphological segmentation method, Morfessor, can help in a supervised setting. Previous research has shown the effectiveness of the approach in semisupervised settings with small amounts of labeled data. The current tasks vary in data size: the amount of word-level annotated training data is much larger, but the amount of sentencelevel annotated training data remains small. Our approach is to pre-segment the input data for a neural sequence-to-sequence model with the unsupervised method. As the unsupervised method can be trained with raw text data, we use Wikipedia to increase the amount of training data. In addition, we train multilingual models for the sentence-level task. The results for the Morfessor-enriched features are mixed, showing benefit for all three sentencelevel tasks but only some of the word-level tasks. The multilingual training yields considerable improvements over the monolingual sentence-level models, but it negates the effect of the enriched features

U2 - 10.18653/v1/2022.sigmorphon-1.16

DO - 10.18653/v1/2022.sigmorphon-1.16

M3 - Conference article in proceedings

SN - 978-1-955917-82-7

SP - 144

EP - 151

BT - Proceedings of the 19th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology

PB - Association for Computational Linguistics

T2 - Meeting of the Special Interest Group on Computational Morphology and Phonology

Y2 - 14 July 2022 through 14 July 2022

ER -

Morfessor-enriched features and multilingual training for canonical morphological segmentation

Abstract

Conference

Access to Document

Fingerprint

Projects

USSEE: Understanding Speech and Scene with Ears and Eyes

Cite this