En-Is Synthetic Parallel Astronomy Corpus with Injected Vocabulary

  Authors

Ingólfsdóttir, Svanhvít Lilja ; Snæbjarnarson, Vésteinn ; Símonarson, Haukur Barri ; Jónsson, Haukur Páll ; Ragnarsson, Pétur Orri ; Þorsteinsson, Vilhjálmur

  Item identifier

http://hdl.handle.net/20.500.12537/135

 Project URL

https://velthyding.is/

 Demo URL

https://velthyding.is/

 Date issued

2021-09-01

 Type

corpus

 Size

20646 sentences

 Language(s)

English , Icelandic

 Description

Synthetic parallel corpus (Icelandic-English) with astronomy terms injected into university abstracts (IPAC, https://arxiv.org/pdf/2108.05289v1.pdf). The data was created using the GreynirTerms utility tool (https://github.com/mideind/GreynirTerms). The sentences are injected with nouns from a glossary of astronomy and physics terms (https://idordabanki.arnastofnun.is/leit//ordabok/STJARNA). The sentences are grammatically correct, as the nouns are injected in the correct grammatical case, gender and number, but semantically they don't make sense. This synthetic data was originally used for training NMT models in order to teach them how to translate rare vocabulary, in fields where true parallel training data is hard to come by. --- Tilbúin samhliða málheild (íslenska-enska) þar sem stjörnufræðihugtökum hefur verið skotið inn í útdrætti háskólaritgerða úr Skemmunni (IPAC, https://arxiv.org/pdf/2108.05289v1.pdf). Gögnin voru búin til með útskiptingatólinu GreynirTerms (https://github.com/mideind/GreynirTerms). Setningarnar innihalda nafnorð tekin úr hugtakalista í stjörnufræði og eðlisfræði (https://idordabanki.arnastofnun.is/leit//ordabok/STJARNA). Setningarnar eru málfræðilega réttar, þar sem nafnorðin koma fyrir í réttri tölu, kyni og falli, en merkingarlega ótækar. Þessi gervigögn voru upprunalega notuð til þess að kenna vélþýðingarlíkönum að þýða sjaldgæfan orðaforða, á sviðum þar sem erfitt er að nálgast samhliða þjálfunargögn.

 Publisher

Miðeind ehf

 Acknowledgement

Ministry of Education, Science and Culture

Project code: V4c – Synthetic corpus

Project name: Language Technology for Icelandic 2019-2023

 Subject(s)

neural machine translation machine translation parallel corpus synthetic corpus

 Collection(s)

Clarin IS

Show full item record