Literature Review Notes | TICO-19

Translation Initiative for COvid-19

With a focus on low-resource languages, TICO-19 aims to make available data in 35 different lanuages (of which 26 lesser resourced languages in addition to the 9 “pivot” languages) for researchers and to assist the development of tools to effectively disseminate critical information for disaster situations such as presented by the current global pandemic.

Dataset development methodology

TICO-19 was created with the following benchmark:

• Sampling from a variety of public sources - such as PubMed, Wikipedia (and related applications), CMU English-Haitian Creole dataset, etc. - covering information domains including but not restricted to: news, testing, travel advisories, scientific development.
• Grouping languages into three main categories:
  • Pivots: resource-rich languages which sometimes serve as an intermediary between English and a target language
  • Priority: 18 languages classified as high-priority by the TWB due to the overwhelming requests they are receiving & “the strategic location of their partners (e.g. the Red Cross)”
  • Important: 8 languages spoken (by millions) in South and South-East Asia
• Ensuring the quality of translation through a two-step human quality control process:
  • Each document is translated by lanaguage service providers (LSP) and edited
  • A selected fraction of the data (and all content from PubMed since it is the hardest to translate) goes through another round of inspection

Developer resources

Additional COVID-19 data are made available in forms of monolingual & parallel corpus gathered from Wikipedia, BBC, VOA, and NGOs, national/state government sources, respectively.

Discussion & future work

• Aside from the baseline results, note that there is a lack of pre-trained MT systems on the low-source languages such as Dari, Myanmar, Oromo, isiZulu, etc. This, along with the difference in BLEU scores between pivot & non-pivot languages highlights the dire need to collect more data for under-represented communities and languages.

• While current MT systems are mostly trained on general-purpose data or particular out-of-domain data, domain adaptation techniques should improve performance. Incorporating terminologies previously unseen in the vocabulary could also help ensure the quality of translation.

• Multilingual NMT systems trained on massive corpus from the web are also promising in improving translation quality for languages in the lower end of data availability.

• Further efforts to increase representation of low-resource lanagues in the public domain corpora in order to prepare for future crises where translation technologies will be needed.

Read more about the TICO-19 effort here.