Our data is released under various Creative Commons licenses.More information
If you love this content, please consider a donation.
Menu
observation: the more translations a sentence has, the more useful each translation is due to indirect links (e.g. an english sentence translated into spanish and french helps spanish people learning french)
problem: boring sentences are easier to translate and thus are more likely to have a lot of translations, so those who translate interesting sentences are making their contributions less useful (in the language pair sense)
proposed solution: make short lists of interesting sentences and get people to focus on those, thus mutually boosting the usefulness of the translations
note: i have *no* idea as to whether or not this is actually a good way of doing things, and especially if it's any better than just using tatominer and looking for translated sentences outside the first page of search results
I don't know... maybe we have that list already. called favorites.
,,actually very good point! thank you!
It's kinda objective what makes an interesting sentence, so making an algorithm for that is hard
Tatominer tries to do something similar, going through words that don't have translations or example sentences
i was thinking of this being a manual thing but now that you bring it up i just *gotta* try automating it
though if i use machine learning for this (and i probably will) i'll have to take care not to include too many sentences with `Tom` in them in the "these are boring" dataset, else it'll latch onto that as the sole indicator of novelty
I mean, how wrong would that be? ;)
Maybe you just exhausted translating that lazy sentences.
You know, the short ones, when you don't even have to sweep your eyes through them, in the first millisecond you know the translation and adding a translation just takes way more time.
If machine learning is a next step move, I have a better idea. Let the machine translate those similar looking, childplay easy, monotonous, time and energy wasting sentences.
a few problems with that unfortunately
for one machines aren't always good at making natural sentences, which is why i try to only use google translate *after* i've already had a stab at doing it myself so that i'm not tempted to accept its version as "good enough"
for two some languages just don't have enough data to be machine-translated, and in fact tatoeba is often used as a data source for them
and for three if we do it on tatoeba and submit them (e.g. for proofreading) we'll have even *more* boring sentences, now in dozens of languages
I agree with Sobz' third point. We should avoid contaminating the corpora of other languages by translating the least interesting sentences.
The best thing to do, in my opinion, is to translate only the sentences that we like the most, so that over time they become a bigger and bigger part of the collection.
We can put those sentences elsewhere. I'm thinking about if we really start translating using machine learning, probably the best we create a separate part for them and not commercialize those sentences as voluntary contributions.
https://tatoeba.org/en/about
"What is Tatoeba?
Tatoeba is a large database of sentences and translations. Its content is ever-growing and results from the voluntary contributions of thousands of members."
This theme comes up again and again and nothing happens.
The ones who said me they hated those sentences said also what else could they do and translated those further til they quit. Well, I don't really understand people. See e.g. al_ex_an_der's sentences. Those are interesting sentences and one can learn of them. No one translates them. Why do you think did he quit Tatoeba? And when one adds a sentence consisting 1,5 words, contibutors consume that like pigs the swill.
So, I'm totally sceptic.
As I see many of the contributors started to add only their own sentences as a l'art pour l'art. This can be the consequence of the boring sentences.
Tatoeba is a very good idea but it seems to me how the world's going, too, and I can't really see much sense of it.
Maybe we are boring.
My last big try to create useful sentences was this list:
Basic Hungarian - useful for tourists
https://tatoeba.org/en/sentences_lists/show/169075
All of the sentences have English translations, because I tried to find translations to them, and still, many of those English sentences have no other translations.
It's 88 sentences from millions.
> It's 88 sentences from millions.
“My life amounts to no more than one drop in a limitless ocean. Yet what is any ocean, but a multitude of drops?” ― David Mitchell, Cloud Atlas
😃
The third page won't open :(
https://tatoeba.org/hu/sentence...075/und?page=3
It open without showing translations:
https://tatoeba.org/en/sentence.../169075?page=3
I get an Internal Error if I want to see it through search.
Showing translations is also very slow.
Yes, I got it opened but only without translations.
> boring sentences are easier to translate
I don't think this is necessarily true.
If you focus on the frequency of individual words, it does seem like it would have to be the case: more common words are likely both boring and easy to translate, while less common words might be interesting, but also harder to translate.
But sentences are not just individual words, but like bags of them, and you can mix and match the contents to balance difficulty and interestingness, e.g. by using mostly common words to keep the sentence easy to translate and then spicing it up with one slightly rarer word to make it more interesting.
My current setup for artisanal mass production of (hopefully) interesting sentences involves a tatominer-style list of not-too-common words which I take inspiration from, plus a frequency-checking script to see whether I accidentally used any even rarer words, and a length limit to prevent me from going full 18th-century novelist with run-on sentences that are so long your eyeballs faint from exhaustion without making it even half the way to the verb at the end of the sentence. I use that to try out a few candidate sentences, and then I add the one I like best.
I'm not sure whether that actually makes me better at writing sentences that are neither too boring nor too hard to translate, because before I adopted this process, my original sentences were mostly only translated by maaster, and after I adopted this process (like, last week, which is why I'm so eager to talk about it) my original sentences are still mostly only getting translated by maaster. Well, at least I myself am not bored.
As you can tell, I was mostly focusing on creating new sentences. But you're right that adding more translations to already translated sentences has additional benefits due to the indirect translations. So trying to make interesting sentences more visible among the sentences we already have is definitely a good idea. I might put my frequency-based bag-of-words model to the task and see whether it comes up with anything... well, interesting.
Suosittelen kääntämään niitä lauseita, joita haluat muidenkin kääntävän, ja vastaavasti kirjoittamaan sellaisia lauseita joita haluat muiden kääntävän. Hankkeen luonteeseen kuuluu, että kaikki saavat kirjoittaa haluamiaan lauseita (tietyissä rajoissa). On hyödyllisempää keskittyä rohkaisemaan hyvän työn tekijöitä tai tarjoamaan heille työkaluja kuin paheksua niitä, joiden valintojen kanssa on eri mieltä.
Erityisesti kannattaa muistaa, että tietokantaa käytetään moniin eri tarkoituksiin. Erilaiset lauseet ovat hyödyllisiä eri tarkoituksissa. Joidenkin kielten kohdalla jokainen, jopa yksinkertaisin, lause on jo suuri saavutus, jos kieli on esimerkiksi erittäin vaarantunut. Myös aloitteleva kielenoppija voi hyötyä yksinkertaisista lauseista. Jos niitä tuntuu olevan liikaa, on tämä suodatuskysymys; tällöin kaivataan työkaluja käyttäjien suodattaa pois tietynlaisia lauseita.
A szöveg (Thanuir) magyarul:
Javaslom, hogy fordítsd le azokat a mondatokat, amelyeket szeretnéd, hogy mások is lefordítsanak, és írd le azokat a mondatokat, amelyeket szeretnéd, hogy mások is lefordítsanak. A projekt jellegéből adódik, hogy mindenki olyan mondatokat írhat, amilyeneket akar (bizonyos határokon belül). Hasznosabb arra összpontosítani, hogy bátorítsuk vagy eszközöket adjunk azoknak, akik jó munkát végeznek, mint hogy helytelenítsük azokat, akiknek a döntéseivel nem értünk egyet.
Érdemes megjegyezni, hogy az adatbázist sokféle célra fogják használni. A különböző kifejezések különböző célokra hasznosak. Egyes nyelvek esetében még a legegyszerűbb mondat is nagy eredmény, ha például a nyelv erősen veszélyeztetett. Még egy kezdő is profitálhat az egyszerű mondatokból. Ha túl soknak tűnik, az szűrési probléma; ebben az esetben olyan eszközökre van szükség, amelyekkel a felhasználók kiszűrhetnek bizonyos típusú mondatokat.
+++
100
https://tatoeba.org/en/sentences_lists/show/169859
This list contains last week's contributions. Was this list compiled automatically, instead of manually? If so, it should be possible to create a similar list containing all the sentences that need translation this week for Tatominer; and if that list is too short, maybe the number of words could be increased just to create a bigger list. Maybe the sentences in that list would be more interesting; although I guess Tatominer only works for some languages.
If you don't want to rely on Tatominer, you could get a list of the 20,000 most common words in the language you want to translate from (most adults know at least that many words in their native language) and pick words at random to translate sentences on Tatoeba, the advantage being that Tatoeba's search engine knows about declension and conjugation.
> Was this list compiled automatically?
Every Saturday, my script identifies added sentences that match the words of Tatominer to automatically update this list and generate the thank you message on the wall.
> a similar list containing all the sentences that need translation this week
If I understand correctly, instead of clicking on each word on Tatominer, you would prefer to browse all the sentences at once. To do this, an easier way would be to add a general link on each Tatominer page like https://tinyurl.com/2r22bjek . I'll see what I can do.
I'm curious to see what such a list would look like, but I don't know if I'll give it much use, because I'm already used to translating 100 sentences a week, one for each word; but maybe Sobsz or other users would be insterested in using a list like that.
Also, maybe the thank-you message should include a link to last week's contributions.
Also also, I'd rather not open TinyURL links...
But thanks for the hard work!
> add a general link on each Tatominer page like https://tinyurl.com/2r22bjek
I share Cangarejo's aversion to link shorteners, but I followed the link anyway and I think that's a good idea.
I've mostly covered the words Tatominer suggests for cmn-deu, and the remaining ones happen to occur mostly in a few sentences each that are a bit tricky to translate. I'm sure I accidentally skipped a few I could manage to translate, but checking each word individually is a bit demoralizing.
For some reason, it never occurred to me to just use an OR query. Well, now that you suggested it, I can just do it myself. https://tatoeba.org/en/sentence...roved=no&user= 641 results, I'm sure I'll find something to translate in there. (And I understand now why you used a link shortener.)
Maybe these sentences should be in random order. Thank you for your time.
You can get that list in random order using the following advanced search.
https://tatoeba.org/sentences/s...59&sort=random
list=169859
sort=random
I know. I'm just saying that maybe it should be the default on Tatominer. Thanks.
It's not possible to set lists to display sentences in a random order at this time. That's why I suggested using the advanced search showing results in a random order.
The button labeled "All Phrases" on the Tatominer page doesn't display a list, it generates an advanced query.
It looked like your comment was added to this comment.
https://tatoeba.org/en/wall/sho...#message_38552
So, I thought you were commenting about having sentences on that list show in a random order.
personally i don't mind it being sorted by shortest first, it means i can go to a page in the middle and it'll have sentences of roughly the length that i like
In that case, the order should be fewest words first, instead of by "relevance".
> Maybe these sentences should be in random order.
Thanks for your comments, but for now, I'm keeping the relevance sort by default to be consistent with the links in the list.
On the topic of sentences with lots of translations, I notice that the most active user on Tatoeba links his sentences to others written in languages that are not listed in his profile. I'm fairly sure he doesn't know all the languages he links to. He does this linking usually after he has added an additional English translation where an existing English translation is already linked to these languages. I suppose the reasoning is this: as the two English sentences have the same meaning, any sentences linked to the original English sentence must also be linked to his new sentence.
But if I'm right, and he can't read those languages, I wonder if it is really acceptable for him to link his sentences to them. The obvious problem with it is that he will reproduce errors in the corpus where existing sentences are linked incorrectly. I would also say that links between languages made by people who don't understand those languages are inherently untrustworthy, and don't serve the purpose of having an accurate corpus.
Does anyone else link their sentences to languages they can't read? Maybe many of you follow the reasoning above, and you do link in such cases. Maybe I'm just being naive again here.
The user in question is an admin, and he's been doing it for years, so you'd be forgiven for thinking that it's acceptable. If the general opinion here is that it's fine, and you all do it, I suppose I might start doing it, too.
I also think it's a problem that a very active contributor with admin status disregards the guidelines of this community. How do you convince other members to follow them then?
Maybe this contributor could explain the reasons for his behavior to us?
Tämä kuulostaa riskialttiilta toiminnalta. Esimerkiksi englannin lyhennykset tai niiden puute, kuten "I am/I'm", kertovat lauseen rekisteristä tai kenties siitä, mitä sanoja painotetaan. Joku toinen kieli saattaa viestiä rekisteriä tai painotusta tavoilla, jotka ovat kauempana toisistaan kuin nämä kaksi, jolloin vain toisen lauseen linkittäminen olisi sopivaa. Tällaisten erojen havaitseminen on vaikeaa, vaikka kieliä osaisi kohtuullisestikin, ja oleellisesti mahdotonta, jos niitä ei osaa lainkaan. Pisteet ja huutomerkin välinen ero vastaavasti.
Toisaalta voi miettiä, että kuinka vakavaa tämä on. Ehkä on arvokkaampaa saada paljon täsmällisesti oikein linkitettyjä lauseita, vaikka mukana olisi muutama hitusen epäilyttävämpi linkki? Harvoin sitä suoran virheen tekee, jos on varovainen, ja sattuu virheitä vaikka osaisikin molempia kieliä.
Kenties olisin tässä taipuvainen luottamaan hyvässä tahdossa tehtyihin linkityksiin ja niiden poistoihin: linkitä yhteen kuuluvat lauseet ja poista virheelliset linkit kun niitä huomaat, ja jos löytyy oikea erimielisyys, aloita siitä keskustelu. Kuitenkin sillä, että ihmiset uskaltavat korjata virheitä ja puutteita kysymättä jatkuvasti lupaa, on myös suuri arvo, ja luultavasti suurempi kuin harvinaista ja hienovaraisten virheiden poistaminen.
Toisaalta toisaalta, jos joku tekee paljon virheitä työtavoillaan, on asiaan syytä puuttua. Mutta ei kai tätä ole havaittu tässä tapauksessa?
> If the general opinion here is that it's fine, and you all do it, I suppose
> I might start doing it, too.
Feel free to do it, or at least experiment with it. That is not something I would do myself, but I would say that it's fine as long as this practice isn't creating incorrect links, or at least not a harmful amount.
As far as I'm aware, we never had a strict rule that contributors should only link sentences in the languages that they know.
If it turns out to be a very problematic behavior, we can always change the source code so that people can only link sentences in languages in their profile. We can also track down which links have been created by a contributor who doesn't know the language they are linking to, and we can remove those links if it's really that bad.
I don't think "blind-linking" should be encouraged because it's much easier to break things than to fix them.
Even if they are rare, bad links are often seen many times before they are removed. In the meantime, every time a learner recognizes a bad link, it makes them question the validity of all the other links.
To build confidence, I think we should leave it to those who know best to validate direct translations.
I fully agree.
For the last ten years, I've seen a lot of bad links created that way. I think it's an existing problem that we shouldn't make worse.
Thank you lbdx, Thanuir, TRANG and marafon for your replies.
I won't be doing any blind-linking. I feel that I have no business linking my sentences to languages I don't understand, regardless.
Let's hope there are enough contributors who have the time and competency in the respective languages to root out the errors introduced, and also the courage to make their objections to these translations known in the comments.
Would you say it's a problem that is bad enough that we should completely prevent people from linking to a language that is not listed in their profile?
Is it a problem that happens with people who freshly became advanced contributors or is it an ongoing issue with all levels of contributors?
I guess it wouldn't solve the problem, given that you can list as many languages as you want. But this solution may reduce the harm and I'd definitely vote for it.
Surprisingly, it mostly concerns the old contributors and even if they have some languages listed in their profile it doesn't mean they master them at the point to do the good linking. The most typical example are the sentences with the formal/informal "you" (vous/tu): some contributors link them quite carelessly.
It may seem too much but I would encourage people to only link the sentences in their native language to the ones in the other languages they know. At least, that's what I try to do myself. It would help us avoid a lot of mistakes.
> Would you say it's a problem that is bad enough that we should completely prevent people from linking to a language that is not listed in their profile?
Yes, I would vote for this.
Other people have made good points about the damage that could be done by blind linking. I want to raise another issue, which is that the benefits that anyone can introduce by blind linking are effectively nonexistent.
If someone with zero knowledge of a language can conclude something about a sentence in that language (namely, that sentence A can be linked to both sentence X and near-duplicate sentence Y), then by definition, that person adds no knowledge, and no value, by inserting the A-Y link. All the knowledge was added by the translator who added the A-X link.
In case anyone wants an example to make this less abstract:
English sentence #3736098 ("I thought perhaps you could help us") was added as an original on January 1, 2015. It was linked by a Macedonian speaker to Macedonian sentence #4168849 ("Мислев дека можеби ќе можеш да ни помогнеш") on May 10, 2015, and then by an Indonesian speaker to Indonesian sentence #5199962 on June 27, 2016. The author of the original sentence then linked it to the effectively duplicate sentence #7173983 ("I thought that perhaps you could help us") on September 18, 2018. Finally, two days ago, he linked both the Macedonian and the Indonesian to the new near-duplicate, despite the fact that neither language is listed in his profile.
It's a mystery to me why he would spend his time with contributions that provide zero value rather than translate from Japanese, a language that he DOES know, into English.
In Russian, the translations should be different:
"I thought perhaps you could help us" - "Я думал, возможно ты сможешь нам помочь"
"I thought that perhaps you could help us" - "Я думал, что возможно ты сможешь нам помочь"
Probably, that also goes for some other languages.
>> It's a mystery to me why he would spend his time with contributions that provide zero value >>
Some people use Tatoeba's sentences for their own projects. I.e. they might need at least an approximate translation to introduce a new English sentence to speakers of another language. Perhaps, the reason is that. For this case, there is a possibility to download the whole Tatoeba database and modify it in any way for someone's personal usage, not affecting Tatoeba itself. I think Tatoeba's owners should provide more information to its users about that.
I wouldn't think it's all so apodictic.
If sentence A has exactly the same meaning as sentence B, one can safely assume that valid translations of A are also valid translations of B.
There is no obvious flaw in this logic. One could even imagine an algorithm that does the linking when someone adds a same-language "translation".
Of course, invalid translations of A are also invalid for B - this would result in a duplication of errors.
The added value may be infinitesimal small, but above zero nonetheless. Imagine sentence A contains a specific term that is used in Northern Germany. Someone adds the same sentence with the Southern equivalent ... and a Turkish sentence now has both as direct translations.
Grammatical modifications (hum, like adding or leaving out "that" in relative clauses) ... well, basically the same argument, but of course it would be enough to just have examples of both ways in different sentences.
Jos kaksi kieltä molemmat sisältävät mahdollisuuden jättää että-sanan pois tietyistä lauseista ilman että merkitys muuttuu selvästi, niin silloin etättömät lauseet ja etälliset lauseet tulisi linkittää toisiinsa, mutta ei ristiin.
I know that he is fishing. - Tiedän että hän on kalastamassa.
I know he is fishing. - Tiedän hänen olevan kalastamassa.
Tällöin molemmille lauseille on tarkempi linkitetty käännös, ja niistä puuttuu vähemmän tarkka käännös, kuin jos kaikki noista olisi linkitetty keskenään.
That's probably the same argument that Selena has brought up for Russian translations.
It would imply a rule that says: Translate as closely as possible and try to mirror the grammar of your source sentence.
German can do a similar thing with relative clauses, and I have the feeling there is a small difference though I can't tell exactly which, so I wouldn't link the two German possibilities.
If a native speaker of English thinks there is no difference between the English alternatives, I would accept crosswise links as equally good.
Yllättyisin jos tässä ei olisi hienovaraisia rekisterieroja.
Yleensä kielenkäyttö, josta jätetään pois ja lyhennetään asioita, on (hivenen tai paljon) epämuodollisempaa. En englannin suhteen osaa sanoa tässä tapauksessa, mutta esimerkiksi akateemisessa kirjoittamisessa suositellaan monien lyhenteiden välttämistä.