En-Is Semi-Synthetic Parallel Name Robustness Corpus

  Authors

Snæbjarnarson, Vésteinn ; Símonarson, Haukur Barri and Þorsteinsson, Vilhjálmur

  Item identifier

http://hdl.handle.net/20.500.12537/74

 Project URL

https://github.com/mideind/GreynirSeq/tree/develop/src/greynirseq/ner

 Date issued

2020-09-28

 Type

corpus, text

 Size

38416 sentences

 Language(s)

English , Icelandic

 Description

A semi-synthetic parallel corpus, based on the ParIce corpus. Person names have been identified in both source and target sentences in each pair. They are then replaced with other names of the same gender and having the same declension, sourced from news articles. By mixing this data with other parallel corpora during training, a network can see many more names than it would otherwise, helping it learn the correct translation of Icelandic person names. Samhliða hálfgervimálheild unnin upp úr ParIce málheildinni. Mannanöfn eru merkt sérstaklega, beggja vegna í hverju setningapari. Þeim er svo skipt út fyrir önnur nöfn af sama kyni og í sama falli. Nöfnin eru fengin af íslenskum fréttavefjum. Með því að blanda þessari málheild saman við aðrar í þjálfun má láta þýðingarlíkan sjá mun fleiri mannanöfn en annars, og hjálpa því þannig að læra meðferð íslenskra mannanafna.

 Publisher

Miðeind ehf.

 Acknowledgement

Ministry of Education, Science and Culture

Project code: Text processing (pre-and postprocessing) (V3b)

Project name: Language Technology for Icelandic 2019-2023

 Subject(s)

machine translation back translation neural machine translation named entity recognition

 Collection(s)

Clarin IS

Show full item record