Improving quality of Language Weaver using a dictionary?

Sabina Fata asked:

A student of mine, who wrote her final dissertation about gender bias in MT, tried the dictionary function in Language Weaver, but did not get good results (the sentences did not adapt to the terms in the dictionary) as compared to DeepL. What would you suggest to improve quality keeping the correct terminology? (She used the generic Language Weaver en-it model)

Parents Reply
  • Dear Paul,

    one example would be the following:

    The two following term has been added to the dictionary:

    Source English "trans"

    Target Italian "persona trans"

    Example sentence 1: In 2019, we surveyed nearly 3,000 trans and nonbinary people from across Canada."

    Italian Jan/Feb 2023: Nel 2019, abbiamo condotto un sondaggio su circa 3.000 persone provenienti da persona trans e di genere non binario provenienti da tutto il Canada

    Italian today: "Nel 2019, abbiamo intervistato circa 3.000 persona trans e persone non binarie provenienti da tutto il Canada."

    Example sentence 2: The closure of the Connect-Clinic is likely to make accessing health care even harder for trans and non-binary people."

    Italian Jan/Feb 23: "La chiusura della Connect-Clinic renderà molto più difficile l'accesso all'assistenza sanitaria per le persone persona trans e di genere non binario"

    Italian today:" La chiusura della Connect-Clinic probabilmente renderà ancora più difficile l'accesso all'assistenza sanitaria per persona trans e persone non binarie" 

    So, there behaviour of the dictionary function in Language Weaver has changed a little bit since Jan/Feb, when Serena wrote her dissertation, but the plural forms are still not recognised.

    She carried out the same tests using DeepL with dictionaries, and the results were slightly better.

    You can find more examples in Serena's dissertation "Gender bias in machine translation: an investigation of the causes, characteristics, and effects of gender-biased MT output" starting from page 92 "De-biasing the system: DeepL glossaries and Language Weaver dictionaries".