Compressed FastText Models for Czech Tagger

D 2022

Compressed FastText Models for Czech Tagger

NEVĚŘILOVÁ, Zuzana

Basic information

Original name

Compressed FastText Models for Czech Tagger

Authors

NEVĚŘILOVÁ, Zuzana (203 Czech Republic, guarantor, belonging to the institution)

Edition

Brno, Proceedings of the Sixteenth Workshop on Recent Advances in Slavonic Natural Languages Processing, RASLAN 2022, p. 79-87, 9 pp. 2022

Publisher

Tribun EU

Other information

Language

English

Type of outcome

Stať ve sborníku

Field of Study

10200 1.2 Computer and information sciences

Country of publisher

Czech Republic

Confidentiality degree

není předmětem státního či obchodního tajemství

Publication form

printed version "print"

References:

Domovská stránka workshopu Plný text

RIV identification code

RIV/00216224:14330/22:00127484

Organization unit

Faculty of Informatics

ISBN

978-80-263-1752-4

ISSN

Keywords in English

model compression; FastText; embedding evaluation; Czech tagger

Změněno: 20/12/2022 12:43, Mgr. Jitka Nováčková

Abstract

V originále

We are building a new tagger for the Czech language that uses two models: the FastText model for word embeddings and a neural network that assigns tags to tokens. In the deployment, we are struggling with model sizes. Since the model size is a common obstacle in various tasks, several compression methods exist. Authors of the methods often claim that the impact on model performance is minimal. However, the evaluation is done on the two tasks the word embeddings are evaluated on: word analogy and word similarity. No information is provided for the evaluation of subsequent tasks. In this paper, we have trained a FastText word embedding model on more recent data. We retrained the tagger with the same parameters using compressed and uncompressed variants of the original FastText model and the new one. After comparing the results, we can see quantization methods are suitable, possibly together with pruning, without significant impact on the tagger performance. The precision dropped by 0.1 percentage point only in quantized models. All tested compression methods reduce the model size 10–100 times.

Links

LM2018101, research and development project

Name: Digitální výzkumná infrastruktura pro jazykové technologie, umění a humanitní vědy (Acronym: LINDAT/CLARIAH-CZ)

Investor: Ministry of Education, Youth and Sports of the CR

Citovat

NEVĚŘILOVÁ, Zuzana. Compressed FastText Models for Czech Tagger. In Aleš Horák, Pavel Rychlý, Adam Rambousek. Proceedings of the Sixteenth Workshop on Recent Advances in Slavonic Natural Languages Processing, RASLAN 2022. Brno: Tribun EU, 2022, p. 79-87. ISBN 978-80-263-1752-4.

@inproceedings{2240154,
   author = {Nevěřilová, Zuzana},
   address = {Brno},
   booktitle = {Proceedings of the Sixteenth Workshop on Recent Advances in Slavonic Natural Languages Processing, RASLAN 2022},
   editor = {Aleš Horák, Pavel Rychlý, Adam Rambousek},
   keywords = {model compression; FastText; embedding evaluation; Czech tagger},
   howpublished = {tištěná verze "print"},
   language = {eng},
   location = {Brno},
   isbn = {978-80-263-1752-4},
   pages = {79-87},
   publisher = {Tribun EU},
   title = {Compressed FastText Models for Czech Tagger},
   url = {https://raslan2022.nlp-consulting.net/},
   year = {2022}
}

TY  - JOUR
ID  - 2240154
AU  - Nevěřilová, Zuzana
PY  - 2022
TI  - Compressed FastText Models for Czech Tagger
PB  - Tribun EU
CY  - Brno
SN  - 9788026317524
KW  - model compression
KW  - FastText
KW  - embedding evaluation
KW  - Czech tagger
UR  - https://raslan2022.nlp-consulting.net/
N2  - We are building a new tagger for the Czech language that uses two models: the FastText model for word embeddings and a neural network that assigns tags to tokens. In the deployment, we are struggling with model sizes. Since the model size is a common obstacle in various tasks, several compression methods exist. Authors of the methods often claim that the impact on model performance is minimal. However, the evaluation is done on the two tasks the word embeddings are evaluated on: word analogy and word similarity. No information is provided for the evaluation of subsequent tasks. In this paper, we have trained a FastText word embedding model on more recent data. We retrained the tagger with the same parameters using compressed and uncompressed variants of the original FastText model and the new one. After comparing the results, we can see quantization methods are suitable, possibly together with pruning, without significant impact on the tagger performance. The precision dropped by 0.1 percentage point only in quantized models. All tested compression methods reduce the model size 10–100 times.
ER  -

NEVĚŘILOVÁ, Zuzana. Compressed FastText Models for Czech Tagger. In Aleš Horák, Pavel Rychlý, Adam Rambousek. \textit{Proceedings of the Sixteenth Workshop on Recent Advances in Slavonic Natural Languages Processing, RASLAN 2022}. Brno: Tribun EU, 2022, p.~79-87. ISBN~978-80-263-1752-4.

Detailed Information on Publication Record