North Sámi morphological segmentation with low-resource semi-supervised sequence labeling

Stig-Arne Grönroos; Sami Virpioja; Mikko Kurimo

North Sámi morphological segmentation with low-resource semi-supervised sequence labeling

Stig-Arne Grönroos, Sami Virpioja, Mikko Kurimo

Speech Recognition

Utopia Analytics Oy

Tutkimustuotos: Artikkeli kirjassa/konferenssijulkaisussa › Conference contribution › Scientific › vertaisarvioitu

75 Lataukset (Pure)

Abstrakti

Semi-supervised sequence labeling is an effective way to train a low-resource
morphological segmentation system. We show that a feature set augmentation
approach, which combines the strengths of generative and discriminative mod-
els, is suitable both for graphical models like conditional random field (CRF) and
sequence-to-sequence neural models. We perform a comparative evaluation be-
tween three existing and one novel semi-supervised segmentation methods. All
four systems are language-independent and have open-source implementations.
We improve on previous best results for North Sámi morphological segmentation.
We see a relative improvement in morph boundary F 1 -score of 8.6% compared
to using the generative Morfessor FlatCat model directly and 2.4% compared to a
seq2seq baseline. Our neural sequence tagging system reaches almost the same
performance as the CRF topline.

Alkuperäiskieli	Englanti
Otsikko	Fifth Workshop on Computational Linguistics for Uralic Languages
Alaotsikko	Proceedings of the Workshop
Kustantaja	Association for Computational Linguistics
Sivut	15-26
Sivumäärä	12
ISBN (elektroninen)	978-1-948087-92-6
Tila	Julkaistu - 7 tammik. 2019
OKM-julkaisutyyppi	A4 Artikkeli konferenssijulkaisuussa
Tapahtuma	International Workshop on Computational Linguistics for Uralic Languages - University of Tartu, Tartu, Viro Kesto: 7 tammik. 2019 → 8 tammik. 2019

Workshop

Workshop	International Workshop on Computational Linguistics for Uralic Languages
Lyhennettä	IWCLUL
Maa/Alue	Viro
Kaupunki	Tartu
Ajanjakso	07/01/2019 → 08/01/2019

Pääsy asiakirjaan

2019_iwclul.publishedFinal published version, 189 KBLisenssi: CC BY-ND

https://www.aclweb.org/anthology/W19-0302/Lisenssi: CC BY-ND

OpenUrl-saatavuus

Koko teksti

MeMAD: Methods for Managing Audiovisual Data: Combining Automatic Efficiency with Human Accuracy
Kurimo, M., Grönroos, S., Brander, T., Porjazovski, D., Raitio, R., Rouhe, A., Grósz, T. & Virkkunen, A.
27/12/2017 → 31/03/2021
Projekti: EU: Framework programmes funding

Science-IT
Mikko Hakala (Manager)
Perustieteiden korkeakoulu
Laitteistot/tilat: Facility

Siteeraa tätä

@inproceedings{832e50deac024e459a9eaf08e5049c1c,

title = "North S{\'a}mi morphological segmentation with low-resource semi-supervised sequence labeling",

abstract = "Semi-supervised sequence labeling is an effective way to train a low-resourcemorphological segmentation system. We show that a feature set augmentationapproach, which combines the strengths of generative and discriminative mod-els, is suitable both for graphical models like conditional random field (CRF) andsequence-to-sequence neural models. We perform a comparative evaluation be-tween three existing and one novel semi-supervised segmentation methods. Allfour systems are language-independent and have open-source implementations.We improve on previous best results for North S{\'a}mi morphological segmentation.We see a relative improvement in morph boundary F 1 -score of 8.6% comparedto using the generative Morfessor FlatCat model directly and 2.4% compared to aseq2seq baseline. Our neural sequence tagging system reaches almost the sameperformance as the CRF topline.",

keywords = "morphology, segmentation, low-resource settings, semi-supervised learning, sequence labeling, recurrent neural networks, conditional random fields, north sami",

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

note = "| openaire: EC/H2020/780069/EU//MeMAD; International Workshop on Computational Linguistics for Uralic Languages, IWCLUL ; Conference date: 07-01-2019 Through 08-01-2019",

year = "2019",

month = jan,

day = "7",

language = "English",

pages = "15--26",

booktitle = "Fifth Workshop on Computational Linguistics for Uralic Languages",

publisher = "Association for Computational Linguistics",

address = "United States",

}

Grönroos, S-A, Virpioja, S & Kurimo, M 2019, North Sámi morphological segmentation with low-resource semi-supervised sequence labeling. julkaisussa Fifth Workshop on Computational Linguistics for Uralic Languages: Proceedings of the Workshop. Association for Computational Linguistics, Sivut 15-26, International Workshop on Computational Linguistics for Uralic Languages, Tartu, Viro, 07/01/2019. <https://www.aclweb.org/anthology/W19-0302/>

North Sámi morphological segmentation with low-resource semi-supervised sequence labeling. / Grönroos, Stig-Arne; Virpioja, Sami; Kurimo, Mikko.
Fifth Workshop on Computational Linguistics for Uralic Languages: Proceedings of the Workshop. Association for Computational Linguistics, 2019. s. 15-26.

Tutkimustuotos: Artikkeli kirjassa/konferenssijulkaisussa › Conference contribution › Scientific › vertaisarvioitu

TY - GEN

T1 - North Sámi morphological segmentation with low-resource semi-supervised sequence labeling

AU - Grönroos, Stig-Arne

AU - Virpioja, Sami

AU - Kurimo, Mikko

N1 - | openaire: EC/H2020/780069/EU//MeMAD

PY - 2019/1/7

Y1 - 2019/1/7

N2 - Semi-supervised sequence labeling is an effective way to train a low-resourcemorphological segmentation system. We show that a feature set augmentationapproach, which combines the strengths of generative and discriminative mod-els, is suitable both for graphical models like conditional random field (CRF) andsequence-to-sequence neural models. We perform a comparative evaluation be-tween three existing and one novel semi-supervised segmentation methods. Allfour systems are language-independent and have open-source implementations.We improve on previous best results for North Sámi morphological segmentation.We see a relative improvement in morph boundary F 1 -score of 8.6% comparedto using the generative Morfessor FlatCat model directly and 2.4% compared to aseq2seq baseline. Our neural sequence tagging system reaches almost the sameperformance as the CRF topline.

AB - Semi-supervised sequence labeling is an effective way to train a low-resourcemorphological segmentation system. We show that a feature set augmentationapproach, which combines the strengths of generative and discriminative mod-els, is suitable both for graphical models like conditional random field (CRF) andsequence-to-sequence neural models. We perform a comparative evaluation be-tween three existing and one novel semi-supervised segmentation methods. Allfour systems are language-independent and have open-source implementations.We improve on previous best results for North Sámi morphological segmentation.We see a relative improvement in morph boundary F 1 -score of 8.6% comparedto using the generative Morfessor FlatCat model directly and 2.4% compared to aseq2seq baseline. Our neural sequence tagging system reaches almost the sameperformance as the CRF topline.

KW - morphology

KW - segmentation

KW - low-resource settings

KW - semi-supervised learning

KW - sequence labeling

KW - recurrent neural networks

KW - conditional random fields

KW - north sami

M3 - Conference contribution

SP - 15

EP - 26

BT - Fifth Workshop on Computational Linguistics for Uralic Languages

PB - Association for Computational Linguistics

T2 - International Workshop on Computational Linguistics for Uralic Languages

Y2 - 7 January 2019 through 8 January 2019

ER -

North Sámi morphological segmentation with low-resource semi-supervised sequence labeling

Abstrakti

Workshop

Pääsy asiakirjaan

OpenUrl-saatavuus

Sormenjälki

Projektit

MeMAD: Methods for Managing Audiovisual Data: Combining Automatic Efficiency with Human Accuracy

Laitteet

Science-IT

Siteeraa tätä