Suppose I am translating the sentence "How are you?" (let is be sentence E1) to Bengali. There would be three possible translations:
B1 - আপনি কেমন আছেন? (Polite)
B2 - তুমি কেমন আছো? (Familiar)
B3 - তুই কেমন আছিস? (Very Familiar)
Now, suppose E1 has also been translated to Spanish, where the possible translations would be (if I am not mistaken):
S1 - ¿Cómo está usted?
S2 - ¿Cómo estás (tú)?
So, B1, B2 and B3 as well as S1 and S2 are directly linked to E1.
Now, suppose someone wants to see the translation of S1 in Bengali. Here is where the problem arises, because all the above mentioned Bengali sentences, i.e B1, B2 and B3 would be shown as the translation, whereas only B1 would be the proper translation of S1.
So my question is, how could this problem be solved? Since I know a bit of Spanish, I could link S1 to B1, but what about other languages which have different pronouns for different level of politeness?
P.S.- The example above has been taken from the "How to be a good contributor" post on the Tatoeba blog.
If the sentences exist and have already been linked correctly, B1, B2, and B3 will all be shown as linked to S1, but only B1 will be shown as directly linked to S1 (with a green arrow), while the other two will be shown as indirectly linked to S1 (with a gray arrow). We tell users that they should only trust direct links.
But I need to link S1 to B1, only then will there be a direct linking, isn't it? But what about, say, Italian which has, I think two forms of "you"? I don't know Italian, so I won't be able to link B1 to any of the Italian sentences. So an Italian sentence in the formal form, linked directly to the E1, would show B1, B2 and B3 as the translations of that Italian sentence, albeit an indirect one.
Actually Italian has three forms of "you": singular informal, singular polite, and plural. Also, there is a special plural polite form which is used rarely now, as I know.
Yes, it's a kind of problem. For example, Russian and Italian both have masculine and feminine forms of adjectives, but English doesn't have them. So, at first, when I sow Italian feminine form as an indirect translation of Russian masculine form (related to English), I considered it as a "mistake". It takes some time to understand, how it works. :)
Something that needs to be made clearer is that by policy, only Directly linked translations are to be treated as 'official' translations. Indirectly linked translations should be trusted about as much as machine translations, possibly even less. Therefore, if B1, B2 and S3 show up, none of these are to be considered a translation of S1 or S2 until a user fluent in both languages constructs links between them.
As for your question about Italian, it must also be kept in mind that it is sound policy to only translate sentences or link them if you are absolutely sure of their meaning. In your hypothetical case, it may be best to leave the sentences be until a user fluent in Bengali and Italian can link sentences as appropriate.
Is it difficult to implement auto-search of duplicates before adding a new sentence?
Stackoverflow has a feature like this: before your question is posted, existing questions are searched by their content for resemblance and if any are found they are displayed for the user.
On Tatoeba, a list of duplicates could be shown for the user with the option to link an existing sentence instead of adding a new one which (s)he has just written.
Sorry if this question was asked here before, I don't know how to search the Wall.
Hello Tatoebans, how is everyone doing?
I have a question regarding translating. I don't know whether a similar question has been answered before, so here I go.
I am currently translating sentences to Bengali. I am a bit confused about sentences containing different grammatical persons because of the fact that Bengali verbs are inflected for person and honour. http://en.wikipedia.org/wiki/Be...grammar#Person
So, while translating a sentence like this:
I think he will be glad to see you. (http://tatoeba.org/eng/sentences/show/71169)
the equivalent Bengali translations would be:
আমার মনে হয় সে তোমাকে দেখে খুশি হবে।
আমার মনে হয় তিনি তোমাকে দেখে খুশি হবেন।
আমার মনে হয় ও তোমাকে দেখে খুশি হবে।
আমার মনে হয় উনি তোমাকে দেখে খুশি হবেন।
This gets even complicated for some other sentences.
So my question is do I add all the possible translations (which could be up to 6 sentences in Bengali)?
yes, you can add every translation you want if you see a sentence that can be said in different ways in another language. i'm doing this with italian. every time i see a sentence that can be said in different ways in italian, i add all the possible versions i come up with
Ok, so here is a more complicated translation. This English sentence (http://tatoeba.org/eng/sentences/show/15832) could be translated in 8 different ways in Bengali, they being different from one another by one or two words!! I have given all the translations, so please have a look and tell whether that is a good thing to do. I am still confused about what to do. :(
I absolutely know that sheepish feeling when you add several translations and wonder if maybe you're doing something wrong. But don't worry, you can link a million translations if the translations themselves are correct.
This sentence i added: http://tatoeba.org/ita/sentences/show/3173766
got manually deleted by another CM just because it was an exact duplicate of another existing sentence. This should be stopped, since there's already a script that automatically does that, even if it's not used very often. Please stop deleting manually exact duplicates, since it's already an automated job that doesn't require the imposition from other users to not add certain sentences just because the script isn't frequently used
The difference between manual and automatic duplicate merging is that an automatic procedure, if written correctly, can preserve the comments and history of all instances of the sentence. It can also preserve audio, if it exists, for one of the sentences. (If and when we allow multiple audio files for a single sentence in the future, it will preserve all the audio.)
The problem is that we don't have a good automatic duplicate merging mechanism right now. The script that was used in the past was run infrequently (perhaps only several times a year). In part, this was because it was a very slow procedure with many restrictions on when it could be run. It also lost audio links. We're in the process of writing new code. I consider this our second highest near-term priority, after restoring access to the wiki and our help/FAQ pages.
In the meantime, there are two "camps": those who believe that contributors should look for existing sentences before adding possible duplicates, and those who believe that this is too burdensome and thus should be a task left to the software, whenever it is ready. I'm not going to comment on the arguments made on both sides, other than to say they both have some validity in my eyes.
However, there is much that can be done to maintain peace in the time before the duplicate merging code is written. It's a good idea for everyone to look for possible duplicates (especially when contributing sentences consisting of common words). On the other hand, I don't see much use in corpus maintainers deleting duplicate sentences. When we have such a high number of duplicate sentences (~25,000 for English is the figure I've seen), any manual deletion will have no real impact. Nor does it seem to prevent people from contributing duplicate sentences. It does, however, succeed in annoying people and leading to arguments, both of which are counterproductive.
Thus I urge everyone to be patient until the duplicate merging code is complete and in use, and to refrain from either contributing duplicate sentences or from deleting the ones that exist.
>Why do you care, if another user delete it or automatical script delete it? What's the difference for you?
The difference is that if native speakers only search for exact duplicates just to delete them, they won't have the time to find mistakes in sentences and post corrections in comments, which, in my opinion, is a more urgent thing to do, rather than removing manually duplicates
I agree with you, but every user have a right to deside, what he/she likes better to do. If someone don't like to post corrections in comments, we can't force him, can we?
I also agree, that quality management system on Tatoeba.org can be improved. I think, every sentence, regardless, if it was created by a native speaker or non native speaker, should be checked, corrected if it's nesessary and tagged OK. Also, I think, it would be better, if we can see, who tagged it. So, if we trust this user, we can trust all the sentences, he/she tagged, regardless, who created those sentences.
Also, I suggest to create a list of users, who are willing to check sentences, in order those users, who are not sure about the quality of their contribution, can ask them.
You can see the user who added a tag. When you move your mouse over the tag, you see ''user: 1234, date: ...''
You just need to add those numbers at the end of this link: http://tatoeba.org/eng/users/show/
but then, we could even miss new translations by doing that. plus, users might actually learn something with those sentences, so it's far from being a waste of time if they translate them, even if they produce duplicates
well, i don't think it's such a problem if users sometimes add some duplicates by chance, so i don't see the need to rush to delete them because the deduplication script isn't used that often. they'll eventually be merged, so if any user decides to add new translations to them, they'll be added to just one sentence after the merging
This change in status isn't really a "promotion" as a reward for good service.
This status change means ...
1. The member can now add tags.
2. The member can add links between existing sentences.
I think any member who has contributed a lot responsibly and feels either of these are things you would like to do to help the corpus, should request this status.
(Perhaps, "a lot" would be over 1,000 sentences.)