@inproceedings{sen-etal-2021-university,
    title = "The {U}niversity of {E}dinburgh`s Submission to the {IWSLT}21 Simultaneous Translation Task",
    author = "Sen, Sukanta  and
      Germann, Ulrich  and
      Haddow, Barry",
    editor = "Federico, Marcello  and
      Waibel, Alex  and
      Costa-juss{\`a}, Marta R.  and
      Niehues, Jan  and
      Stuker, Sebastian  and
      Salesky, Elizabeth",
    booktitle = "Proceedings of the 18th International Conference on Spoken Language Translation (IWSLT 2021)",
    month = aug,
    year = "2021",
    address = "Bangkok, Thailand (online)",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/Ingest-2025-COMPUTEL/2021.iwslt-1.4/",
    doi = "10.18653/v1/2021.iwslt-1.4",
    pages = "46--51",
    abstract = "We describe our submission to the IWSLT 2021 shared task on simultaneous text-to-text English-German translation. Our system is based on the re-translation approach where the agent re-translates the whole source prefix each time it receives a new source token. This approach has the advantage of being able to use a standard neural machine translation (NMT) inference engine with beam search, however, there is a risk that incompatibility between successive re-translations will degrade the output. To improve the quality of the translations, we experiment with various approaches: we use a fixed size wait at the beginning of the sentence, we use a language model score to detect translatable units, and we apply dynamic masking to determine when the translation is unstable. We find that a combination of dynamic masking and language model score obtains the best latency-quality trade-off."
}