Efficient Architectures For Low-Resource Machine Translation

D 2025

Efficient Architectures For Low-Resource Machine Translation

SIGNORONI, Edoardo; Pavel RYCHLÝ a Ruggero SIGNORONI

Základní údaje

Originální název

Efficient Architectures For Low-Resource Machine Translation

Autoři

SIGNORONI, Edoardo ; Pavel RYCHLÝ a Ruggero SIGNORONI

Vydání

Shoumen, BULGARIA, Proceedings of the First Workshop on Advancing NLP for Low-Resource Languages associated with the International Conference RANLP 2025, od s. 39-64, 26 s. 2025

Nakladatel

INCOMA Ltd.

Další údaje

Jazyk

angličtina

Typ výsledku

Stať ve sborníku

Obor

10200 1.2 Computer and information sciences

Stát vydavatele

Bulharsko

Utajení

není předmětem státního či obchodního tajemství

Forma vydání

elektronická verze "online"

Odkazy

URL

Označené pro přenos do RIV

Ano

Kód RIV

RIV/00216224:14330/25:00143189

Organizační jednotka

Fakulta informatiky

ISBN

978-954-452-100-4

Klíčová slova anglicky

Machine Translation; Low-Resource Languages

Příznaky

Mezinárodní význam, Recenzováno

Změněno: 17. 3. 2026 14:06, Edoardo Signoroni

Anotace

V originále

Low-resource Neural Machine Translation is highly sensitive to hyperparameters and needs careful tuning to achieve the best results with small amounts of training data. We focus on exploring the impact of changes in the Transformer architecture on downstream translation quality, and propose a metric to score the computational efficiency of such changes. By experimenting on English-Akkadian, German-Lower Sorbian, English-Italian, and English-Manipuri, we confirm previous finding in low-resource machine translation optimization, and show that smaller and more parameter-efficient models can achieve the same translation quality of larger and unwieldy ones at a fraction of the computational cost. Optimized models have around 95% less parameters, while dropping only up to 14.8% ChrF. We compile a list of optimal ranges for each hyperparameter.

Návaznosti

LM2023062, projekt VaV

Název: Digitální výzkumná infrastruktura pro jazykové technologie, umění a humanitní vědy

Investor: Ministerstvo školství, mládeže a tělovýchovy ČR, LINDAT/CLARIAH-CZ - Digitální výzkumná infrastruktura pro jazykové technologie, umění a humanitní vědy

Přiložené soubory

2025.lowresnlp-1.6.pdf Verze souboru

Citovat

SIGNORONI, Edoardo; Pavel RYCHLÝ a Ruggero SIGNORONI. Efficient Architectures For Low-Resource Machine Translation. Online. In Ernesto Luis Estevanell-Valladares, Alicia Picazo-Izquierdo, Tharindu Ranasinghe, Besik Mikaberidze, Simon Ostermann, Daniil Gurgurov, Philipp Müller, Kurt Micallef, Claudia Borg, Michal Gregor and Marián Šimko. Proceedings of the First Workshop on Advancing NLP for Low-Resource Languages associated with the International Conference RANLP 2025. Shoumen, BULGARIA: INCOMA Ltd., 2025, s. 39-64. ISBN 978-954-452-100-4.

@inproceedings{2545110,
   author = {Signoroni, Edoardo and Rychlý, Pavel and Signoroni, Ruggero},
   address = {Shoumen, BULGARIA},
   booktitle = {Proceedings of the First Workshop on Advancing NLP for Low-Resource Languages associated with the International Conference RANLP 2025},
   editor = {Ernesto Luis Estevanell-Valladares, Alicia Picazo-Izquierdo, Tharindu Ranasinghe, Besik Mikaberidze, Simon Ostermann, Daniil Gurgurov, Philipp Müller, Kurt Micallef, Claudia Borg, Michal Gregor and Marián Šimko},
   keywords = {Machine Translation; Low-Resource Languages},
   howpublished = {elektronická verze "online"},
   language = {eng},
   location = {Shoumen, BULGARIA},
   isbn = {978-954-452-100-4},
   pages = {39-64},
   publisher = {INCOMA Ltd.},
   title = {Efficient Architectures For Low-Resource Machine Translation},
   url = {https://acl-bg.org/proceedings/2025/LowResNLP%202025/pdf/2025.lowresnlp-1.6.pdf},
   year = {2025}
}

TY  - CONF
ID  - 2545110
AU  - Signoroni, Edoardo - Rychlý, Pavel - Signoroni, Ruggero
PY  - 2025
TI  - Efficient Architectures For Low-Resource Machine Translation
PB  - INCOMA Ltd.
CY  - Shoumen, BULGARIA
SN  - 9789544521004
KW  - Machine Translation
KW  - Low-Resource Languages
UR  - https://acl-bg.org/proceedings/2025/LowResNLP%202025/pdf/2025.lowresnlp-1.6.pdf
N2  - Low-resource Neural Machine Translation is highly sensitive to hyperparameters and needs careful tuning to achieve the best results with small amounts of training data. We focus on exploring the impact of changes in the Transformer architecture on downstream translation quality, and propose a metric to score the computational efficiency of such changes. By experimenting on English-Akkadian, German-Lower Sorbian, English-Italian, and English-Manipuri, we confirm previous finding in low-resource machine translation optimization, and show that smaller and more parameter-efficient models can achieve the same translation quality of larger and unwieldy ones at a fraction of the computational cost. Optimized models have around 95% less parameters, while dropping only up to 14.8% ChrF. We compile a list of optimal ranges for each hyperparameter.
ER  -

SIGNORONI, Edoardo; Pavel RYCHLÝ a Ruggero SIGNORONI. Efficient Architectures For Low-Resource Machine Translation. Online. In Ernesto Luis Estevanell-Valladares, Alicia Picazo-Izquierdo, Tharindu Ranasinghe, Besik Mikaberidze, Simon Ostermann, Daniil Gurgurov, Philipp Müller, Kurt Micallef, Claudia Borg, Michal Gregor and Marián Šimko. \textit{Proceedings of the First Workshop on Advancing NLP for Low-Resource Languages associated with the International Conference RANLP 2025}. Shoumen, BULGARIA: INCOMA Ltd., 2025, s.~39-64. ISBN~978-954-452-100-4.

Přehled o publikaci