Duolingo is the most popular way to learn languages in the world. Best of all, it's 100% free!

https://www.duolingo.com/Vinotinto

The translation computer program

Is DL attempting to teach the computer translation program English?

5 years ago

19 Comments


https://www.duolingo.com/fuonk
fuonk
  • 25
  • 14
  • 2

I agree with Vinotinto that there are several ways in which DuoLingo is not as good a learning environment for us humans as the DuoLingo people suggest it is.

It would be nice if at least the process were producing a more capable translation engine, but I haven't seen or heard anything which suggests that the translation engines they are using are actually being trained by the process.

The lessons are being improved by the alternative translations users send in when they are told their answers are wrong when doing exercises. That is presumably not an automated process, however-- human beings are evaluating the alternative translations which are sent in. Meanwhile, the exercises train us to produce translations we think are more likely to be accepted rather than translations we think are natural.

5 years ago

https://www.duolingo.com/DonnaMarie

They might want to do that but after working on these translations for a few weeks I see how much the human brain needs to be involved. You need to make judgements about the meaning, come up with similar words that might not be in the Duolingo database and adapt to the modern use of English.

5 years ago

https://www.duolingo.com/Vinotinto

@ DonnaMarie: Thanks for the comment, However, An English speaking Person needs to be involved in determining a correct translation. What we have is the blind leading the blind leading the blind off of the edge of a cliff, with the computer at the head of the line. This is a poorly designed learning tool...except possibly for the computer. Even that is questionable.

5 years ago

https://www.duolingo.com/fuonk
fuonk
  • 25
  • 14
  • 2

No. DuoLingo is not developing any new translation software. They are experimenting with a process which combines the translations produced by existing machine translation programs, which are known to be bad, with translations produced by us language learners. Their claim is that by cleverly combining these translations and using the judgements of learners about what translations sound correct, they can produce translations which are equal in quality to those produced by professional translators.

Personally, I am skeptical about this claim. It is definitely true that in most cases the process results in a better translation than any of the individual contributors could have made, but I have seen many "final results" which still contain significant errors.

5 years ago

https://www.duolingo.com/fuonk
fuonk
  • 25
  • 14
  • 2

I think they are testing two hypotheses. One is that students will learn the languages they are studying simply by engaging in the process of translating "real-world" sentences and seeing how other people translate the same sentences. The other is that "professional quality" translations can be produced by a process involving only machine translations and translations produced by novice translators (language learners). If they can claim that both of these hypotheses are correct, then they can offer the learning experience in exchange for putting the students to work doing a job which would normally have to be done by a highly trained expert.

I don't see any indication that anyone at DuoLingo has any ideas about educating students beyond the process you already see built into the site. The whole point of it from an economic point of view is that it has to be automated; if they have to employ people as ongoing educators to participate in the program, the whole rationale of the program is undermined.

If the process is capable of producing adequate (salable) translations, but does not result in much improvement in the students' abilities, then it is exploitative. If it is incapable of producing adequate translations, then it is of no commercial interest. The study of the "effectiveness" of the site in improving the language skills of people who use it is designed to produce an answer to the first question (I think it is designed to produce an affirmative answer, but that has to do with my skepticism about how the "experiment" has been designed.) I am a participant in that study, but I intend to tell the people who were hired to do the study very clearly what I consider weaknesses both in the design of DuoLingo and in the design of the study.

5 years ago

https://www.duolingo.com/Vinotinto

@fuonk: Thank you for explaining the thoughts behind the process. I think I see the problem causing students to object so strongly. The problem is two-fold. One is the nature of education in the West. Two is the expectations of students. We need, at least at some level, a way to feel accomplishment. The translation proscess as it stands frustrates students. Perhaps DL should bridge the gap and provide a graduated translation process which leads to a student being able to accuratly translate documents? I know document translation is a very specialized area of study. This seems to be what we are being asked to do. Thank you.

5 years ago

https://www.duolingo.com/Vinotinto

Thank you fuonk. The hypothesis is good. And, the translation engine is needed to continue to grow DL. However, the weakness at this time, is the students linguistic ability vs. the computers linguistic ability. I think what DL wants to do can be accomplished. I just think DL needs to start at a more realistic level. Thank you again very much for enlightening me to the reasoning behind the "Real World" translation proigrams. Blessings, Vinotinto.

5 years ago

https://www.duolingo.com/fuonk
fuonk
  • 25
  • 14
  • 2

Actually, Professor Vesselinov responded quite politely when I contacted him on several occasions. In contrast, I have not had a single response from anyone at DuoLingo. Prof. Vesselinov, as he said, doesn't work for DuoLingo; he is part of an independent group hired to determine whether DuoLingo is an effective teaching tool. I think you may have misunderstood what he said about "statistically significant results". Someone doing an experiment always want to get statistically significant results-- that is, results from which one can draw a conclusion. That does not mean, as you seem to be taking it to mean, that he wants the results to "turn out one way or the other". A conclusion that the site is not contributing effectively to learning could also be a statistically significant result. That having been said, I have my own doubts concerning whether the design of the study is adequate to produce meaningful results.

5 years ago

https://www.duolingo.com/territurtle

@Vinotino -- I knew all the grammar and vocabulary in the exercises before doing them. So I found the exercises extremely helpful. Mainly, because I was a sloppy reader. Every time DuoLingo caught me in not paying attention to noun/adjective agreement and other such things, it helped to sharpen my skills. However, I was always aware of the failings of good old DuoBot.

5 years ago

https://www.duolingo.com/Vinotinto

@ territurtle: Narcissis and Goldman (Hesse).....I have yet to see a satisfactory translation. So, I accept your challenge. Please show me an example of one of your translations. And, please explain if it is in American English or in the Queen's English. You know the computer will never have the ability to distinguish between the two, which is one reason for unsatisfactory performance.

5 years ago

https://www.duolingo.com/DonnaMarie

Vinotinto: I don't think you understood my comment. I agree a human brain that has knowledge of the English language is still needed. And I wasn't thinking of the Queen's English! Still trying to figure out why they call a race car driver a "pilot". But that is what is in the database. Everyone just types that in. Some knowledge of the subject matter really is needed as well.

5 years ago

https://www.duolingo.com/rspreng

@fuonk I am also in that study. I emailed the guy in charge, from whom we get the emails, and told him that I was spending a significant amount of time not learning Spanish, but dealing with DL's oddities. He said he didn't care. He was simply hired to do the study, and all he was hoping for was "statistically significant results." :)

5 years ago

https://www.duolingo.com/DonnaMarie

Well they can also have statisically signifiant results that show it doesn't work. Sounds like a flip response but as a researcher, he really shouldn't be invested in the results turning out one way or the other. I guess that is why they don't want to hear our opinions.

5 years ago

https://www.duolingo.com/rspreng

Yep, DonnaMarie, I agree. The researcher doesn't care, but he said he'd pass my message on to the folks at DL. I would think that the DL people would care if they funded the study, but . . . . . ;)

5 years ago

https://www.duolingo.com/Vinotinto

@ all: if nothing else, this is a fantastic discussion. I think I am learning more from the discussion than from DL.

5 years ago

https://www.duolingo.com/DonnaMarie

I'm going to quit complaining and just work on the Skill Tree.

5 years ago

https://www.duolingo.com/Vinotinto

@DonnaMarie: Me too.

5 years ago

https://www.duolingo.com/rspreng

Hey guys! Constant complaining is what distinguishes humans from "lower" animals! ;)

5 years ago

https://www.duolingo.com/DonnaMarie

Glad to hear we are so distinguished!

5 years ago