User talk:Muki987
Contents
Jimregan's remarks
- 'With this knowledge we can construct the English' -- How? You don't seem to have given thought to that part.
- 'háza (his, her, its house repeat all previous to this) - 56' -- it strikes me as a) unlikely that you can chain all possible possessives in this manner and b) that you can do something useful that will convey an understandable meaning in another language even if it is.
- 'házas (married- repeat all previous for this up to here, except the last 2) 1680' -- a married house? Really?
- 'házacska' -- are there no lexicalised diminutives in Hungarian? I can theoretically add '-let' to any noun in English, but 'piglet' has a separate translation to most languages, and 'hamlet' is not a diminutive of 'ham'.
- Just because you can theoretically infer meaning from an analysis doesn't mean that results will translate. -- Jimregan 05:24, 7 April 2009 (UTC)
To Jimregan
>'With this knowledge we can construct the English' -- How? You don't seem to have given thought to that part.
Of course I did. Whatever I can do as a human translator, the machine can also do, if I tell him how. I am absolutely optimistic in the fact, and looking for the proper technology.
- Great. 'To a person with a hammer, all problems look like nails'. I say that Hunmorph is your hammer; you are mixing derivational processes with agglutination and 'normal' morphology. Just because all of these things can be treated the same doesn't mean it always makes sense to do so, which is the point underlying everything I said. -- Jimregan 12:57, 7 April 2009 (UTC)
- Your pont remains unclear for me, but it might be not worth to seek for clearness in this case, since your text seems to be philosophical for me. I am a practicing person, less philosophic type. I am rather new to practicing Hunmorph, anyway. Muki987 18:16, 7 April 2009 (UTC)
- My point is that you have one solution to one problem; you're trying to use that solution for other problems. Clear? -- Jimregan 10:46, 8 April 2009 (UTC)
- Sorry, not. Please explain what you want to say more detailed with examples. Also explain, why are you saying that. Muki987 11:49, 8 April 2009 (UTC)
- I'll make it as simple as possible: you think there's only one problem; there are many more. You are ignoring them because you have one solution, and think it will work for them all. It won't -- Jimregan 15:52, 8 April 2009 (UTC)
- You misunderstand me completely. I very clearly see, we have lots of things to do, I just address one of them, that's all. The main one for me at the moment. If that is fixed, I'll continue with the rest, or even better, we have a lot of commonly solvable problems, and we solve together the rest. What I addressed, is no problem for prefix type language pairs, but very clearly a problem for me. My primer focus is English-Hungarian, German-Hungarian, second English-German, third German-English, Fourth Hungarian-German, Hungarian-English. The other option is, we say, it is impossible to write a translator, I think, that is simply wrong. Muki987 18:14, 8 April 2009 (UTC)
- If I misunderstood you; good. Because it seemed clear to me that you were ignoring other issues, thinking that they would all be solved by using HunMorph. I'd much rather make you angry now than see you work for months, only to find you have to redo everything. For English-Hungarian, you will most likely find it easier to treat certain types of words - derivatives - as separate words, rather than forms of a base word. It's still possible to do otherwise, but it causes a lot of unnecessary complication, and will have undesired side-effects. I find it hard to believe that Hungarian-German will be much different. (Or, honestly, anything other than Hungarian-Finnish and Hungarian-Estonian).
- Yes,I agree with all that. Especially with the remark, German-Hungarian relation compared to English-German relation. I found while translating lots of texts from both, I was faced almost all the time with the same problems. The only exception is in English the lots of meanings of the same word, that fortunately is not the case for German or Hungarian. In English-German relation the only grammatical problem I saw, (besides using of the false words or expressions, which is a general problem for any language pair) is the position of verbs, that tend to be often at the end of a structure, while in English in the middle of it.
- Example:
- Er dachte, sie würde in die Schule gehen.
- He thought, she would go to school.
- Er dachte, sie würde gehen in die Schule --is unusual in German, and sounds un-German. One can see such bad structures in swallow translated texts.
- I believe, Apertium has standard tools to fix this. I also think, to handle words near to the stem is simpler, than try to complicate our life with derivatives (Ház-házas- házas can, and should clearly be considered az an independent word.). My example was just set up to illustrate the great number of derivatives (I forgot even some), and not to suggest any special way to translate that word. Muki987 09:31, 9 April 2009 (UTC)
- Unfortunately, we don't have an effective way of dealing with this example -- but, we (well, Francis and I) recently learned something about our transfer architecture that can possibly be used to deal with this (n-level transfer), but neither of us have had a chance to experiment with it yet. (At least, I haven't; perhaps Francis has). -- Jimregan 10:58, 9 April 2009 (UTC)
- Word order is very important, not only in English-German, but also in German-Hungarian relation.
- He thought, she would go to school.
- Azt hitte, iskolába megy. (azt=that, hitte=thought, iskolába=to school, megy go), when we want to say, she goes to school, and not elsewhere.
- Azt hitte, megy az iskolába- when we want to say, she went and did not fly or swim).
- This also exists in German, we put to the begin the part, we want to stress.
- In die Schule wollte sie gehen. (To school she wanted to go)
- Gehen wollte sie in die Schule. (She wanted to go to school) Muki987 11:18, 9 April 2009 (UTC)
- Have a look at apertium-eu-es (Basque->Spanish). It's one direction only, but using HunMorph would limit you to only being able to translate from Hungarian anyway. (AFAIK, the main reason eu-es is one direction only is because Matxin already exists for the other direction) -- Jimregan 09:01, 9 April 2009 (UTC)
- Yes, I am doing that at the moment, thanks. My priority is, as you know, English-Hungarian, German-Hungarian first. Muki987 09:31, 9 April 2009 (UTC)
>'háza (his, her, its house repeat all previous to this) - 56' -- it strikes me as a) unlikely that you can chain all possible possessives in this manner and b) that you can do something useful that will convey an understandable meaning in another language even if it is.
ház- házam, házad, háza, házunk, házatok házuk (my house, your house, his, her its house, our house, your hous their house) All relations to MY HOUSE are then expressed, as in the case of ház: házban- házamban házra- házamra etc... It is simple and understandable in all cultur languages.
- You aren't addressing my point. 'repeat all previous to this', implying that you can have some combination meaning 'my your his their house'. -- Jimregan 12:57, 7 April 2009 (UTC)
- Repeat all previus means, that I can express the relations to "my house" "your house" ... "Their house" by using exactly the same inflects, as for "house". Above the example with "ban" = in , all others work exactly on the same way. Muki987 18:16, 7 April 2009 (UTC)
- So, referring just to grammatical cases? Ok, that answers my question -- Jimregan 10:46, 8 April 2009 (UTC)
- If you want to express it like that. Neither English, nor Hungarian have grammatical cases in fact, just to be precise. Muki987 11:56, 8 April 2009 (UTC)
- Hungarian does have grammatical cases; 'I usually quote 17 following those established by Antal László in 1977' -- the first group in your set of examples are grammatical cases -- Jimregan 15:52, 8 April 2009 (UTC)
- If I was in you, I were much more modest in my statements. Antal László's linguistic ideas are disputable. In Hungary, nobody speaks about n cases, because that is simply contraproductive. It is also contraproductive for foreigners, if they learn Hungarian. I see now, it is useful for translation, so I will use the concept, but for this purpose only. Muki987 18:14, 8 April 2009 (UTC)
- You misunderstand me here; I was not being immodest: all of the literature I could find in English is in agreement. It may be the case that the views are disputed, but that is not represented in English writings about Hungarian grammar, as far as I have seen at least. How are they considered, then? Because it may be the case that it could be easier to translate to and from Hungarian if the set of suffixes I would regard as case endings were instead treated as enclitic postfixes (that is, by splitting off the suffix and treating it as if it was a separate word: see, for example, how 'dímelo' in Spanish is split into 'decir<vblex><imp><p2><sg>+prpers<prn><enc><p1><mf><sg>+lo<prn><enc><p3><nt>'. Does Hungarian have vowel harmony, like Finnish? That may complicate things, but I think there's a (relatively) easy way around it. -- Jimregan 09:01, 9 April 2009 (UTC)
- I see. You consider English literature as authoritativ for Hungarian grammar questions? I would not do that. It is written in fact by analphabets from the linguistic point of view. Authoritative are in my opinion are only mother tongue authors, who agree with most of the other Hungarian mother tongue linguists. (of course, this includes linguists of the past also, not only at present).
- No, no; I don't speak Hungarian, so I can't check the literature in Hungarian: I have to rely on literature in English. -- Jimregan 10:52, 9 April 2009 (UTC)
- You are however, IMHO absolutely correct, if you say, we must classify postfixes for translation purposes, since we MUST find a way to match our postfixes to the prepositions, and therefore we must classify them. We can call the classification anything, IMHO the best name is classification, but we can call them also cases, which has very little to do with German type cases.
- Great. See Francis' example, below: this is basically what I propose, for analysis. For generation, I propose taking that 'pseudo word' system, and converting it into a string of tags, much like HunMorph generates, in a set of Hungarian-only rules (most of our rules are based on knowledge of both languages, but this set could be reused among language pairs using Hungarian). -- Jimregan 10:52, 9 April 2009 (UTC)
- Yes we do have vowel harmony, however, this is almost trivial: high words get high ending, low low endings, there are a few exceptions, that can be handled by rules. eéiíöüõû are high, aáouóú are low vocales. Every postfix has low and high form, for example ba, be (into) ajtó-ba, szék-be. If the word is mixed, (mixed are typically words taken over from foreign languages) for example radio, either the last syllable decides or we use low, for example rádió-ba. Exceptions are some ancient words for etymological reasons, for example derék, derékba, íj, íjba, they are only a handful words, no problem. Muki987 09:31, 9 April 2009 (UTC)
- Well, exceptions are exceptions, and every language has them. You're right; vowel harmony is not a big problem (at least, not in my opinion) -- but it does mean I need to ask you for more examples, as one set of suffixes is not enough. I know I can take them from Hunspell/Hunmorph (in fact, it would be a requirement IMO, to be able to reuse that data as quickly and easily as possible), but I'd rather focus on one small dataset to begin with, and expand later -- Jimregan 10:52, 9 April 2009 (UTC)
Basque
This conversation is a bit heated for me, but note that in our Basque→Spanish system we do something similar with Basque cases. For example, a typical way of representing "hegoak" would be:
^hegoak/hego<n>+a<det><art><pl>/ hego<n>+a<det><art><sg>+k<post>$
"<hegoak>" "hego" IZE ARR DEK ABS NUMP MUGM "hego" IZE ARR DEK ERG NUMS MUGM
Note how in our representation the case is marked as a postfix k<post>
where in the more traditional analysis it is marked as a case ERG (ergative). Compared with Basque→Spanish, Hungarian→English would be easier in terms of word order:
S O V Txinako Poliziak, datu ofizialen arabera, 1.317 pertsona atzeman zituen la Policía de China, según los datos oficiales, 1.317 personas capturó The Chinese police, according to official data, 1,317 people detained. `According to official data the Chinese police detained 1,317 people.'
- Francis Tyers 10:13, 9 April 2009 (UTC)
- I think, the discussion gets calmer, since I start to see the pont of JimRegan, (and he mine) which is helpful and valid. I see, Matxin can handle both Basque-Spanish and Spanish-Basque, so I'll look throughoutly into that. Basque is also a hun language, as far as I know, very similar to Hungarian. Muki987 10:32, 9 April 2009 (UTC)
- Actually, the Matxin system cannot handle Basque→Spanish, as there is no dependency analysis for Basque. Apertium is used for Basque→Spanish and Matxin for Spanish→Basque. As far as I know, Basque does not have any living relatives. - Francis Tyers 10:44, 9 April 2009 (UTC)
- That's important for me to know, thanks. Muki987 11:30, 9 April 2009 (UTC)
- Is there any difference between the main diagram "How Apertium works" between Apertium and Matxin? If yes, where, if not: What is the difference between Matxin and Apertium (except of character coding)? Muki987 12:37, 9 April 2009 (UTC)
- Basque has lots of living relatives, Hungarian, Armenian, Turkish, Aserbaidshan, Uigur, Finnish, Estonian, Persian, Japanese (thru Ainu = hunnish influence), Ketchua (Inka language in south America), ancient Egyptian (no more living, but hieroglyphes show a great past), Etruscian (also no more living, but great past), Hindi, and more. Muki987 11:30, 9 April 2009 (UTC)
- I do not agree, although the issue is not really pertinent to our current discussion. - Francis Tyers 11:53, 9 April 2009 (UTC)
Considerations for prefix groups and possessions
Prefix groups
In English, one prefix can handle more nouns, for example: I travel to England, France and Spain. This will be translated as: Utazom Angliá-ba, Franciaország-ba és Spanyolország-ba. ("-" added for clarification). Utazom: I travel, Angliába: to England, ... , Spanyolországba: to Spain
In English the prefix ...nouns structure will be closed by:
- a dot (finishing the sentence)
- a verb - I travel to england an spain and will carry a bag- the word "will" closes the scope of to.
- a new prefix - I travel to england and spain with train or aeroplane- the word "with" closes the scope of to.
- Co-ordinated noun phrase with case agreement. I would probably do this kind of thing in pre-transfer with a constraint grammar. Basically write a rule which does: "add accusative case to nouns following the preposition 'to' until a new preposition, verb or end-of-sentence". - Francis Tyers 10:22, 9 April 2009 (UTC)
Possesions
In English the possessor may be before the possesion: Peter's coffee and tee
but also behind it: the coffee and tee of Peter
In Hungarian the possessor is always strictly before the possesion, both sentences above must be translated as: Péter kávé-ja és teá-ja. (again "-" just for clarity).
In English the possession structure will be closed by
- a dot (finishing the sentence)
- a verb - Peters coffee and tee looks like a bag - the word "looks" closes the scope of possession structure
- a new prefix - Peters coffee and tee with sugar - the word "with" closes the scope of possession structure
In the case of "the coffee and tee of Peter" type possession relation: If an noun enumeration starts, the translator must watch. If the enumeration ends with "of", this is a possession structure, and must be translated, os such.
Combination of possession and prefix
- With Peter's coffee and tee - Péter kávé-já-val és teá-já-val - ja is possession, val, vel is with
- With the coffe and tee of Peter - as above
Adding plural
- With Peter's coffee and tees - Péter kávé-já-val és teá-i-val - "i" is plural possession for tea
- Oh. That's interesting, that plurality 'goes with' the possessive. Not really an extra problem, but it is interesting. -- Jimregan 11:31, 9 April 2009 (UTC)
Remark
These kind of structures caused for me the most manual work when translated texts from English/German, therefore it is very important to set up their proper translation. Thanks in advance for any critics/thought/comments. Muki987 10:12, 9 April 2009 (UTC)
- Yes; they pose quite a problem, because the phrase boundary needs to be detected: in 'Peter's coffee and tea', 'coffee and tea' is the part that's possessed, but in the sentence 'I drank Peter's coffee and tea was spilled on the ground' only 'coffee' is possessed. We can use CG to add boundaries here, but it will be a lot of work. -- Jimregan 11:29, 9 April 2009 (UTC)
- An interesting example. This is a fourth kind of structure closing signal: noun immediately followed a verb also stops the structure:
- I drank Peter's coffee and children played near to us.
- I saw Peter's coffee and tee smell like sugar - this sentence is even in English is ambiquous - what smells like sugar, both or only tee? Would a comma after coffee limit possession to coffee? Muki987 11:48, 9 April 2009 (UTC)
- I'm not sure if this is ambiguous. The ambiguity is resolved by the inflection of the verb in this particular case.
- ?I saw Peter's coffee and tee smell like sugar
- I saw [Peter's coffee] and [tea] smell like sugar
- I saw [Peter's coffee and tea] smell like sugar
- - Francis Tyers 13:36, 9 April 2009 (UTC)
- This particular example aside, the premise is sound: English has ambiguities that can cause difficulties in determining phrase boundaries. -- Jimregan 14:13, 9 April 2009 (UTC)
- Indeed. - Francis Tyers 14:23, 9 April 2009 (UTC)
- Yes. Apertium's transfer works on left to right longest match. We were hoping that someone would be interested in integrating CG's dependency analysis for GSoC, which would help to resolve these ambiguities; at the moment, we have to simply pick the most common cases, and fail in others. -- Jimregan 13:32, 9 April 2009 (UTC)
Some tests
Peter's coffee and tee are sweet
- el café de Peter y tee es dulce
Peter's mice and cats are sweet
- los ratones de Peter y los gatos son dulces
Peter's mice and Peter's cats are sweet
- los ratones de Peter y los gatos de Peter son dulces
This clearly shows, that in case of possesion apertium would not consider multiple possessions. Since it is also ambiguious in English, this is a feature.
- Peter's coffee and tea are sweet
- el café de Peter y el té son dulces
- In this case, rules are not available for co-ordinated noun phrases, and so the translations come out rather badly. It is not to say that it is impossible to do in Apertium, just that so far for English→Spanish we have had more important things to work on. - Francis Tyers 13:27, 10 April 2009 (UTC)
Same is true for attributes:
Little boys and girls were playing
- Pocos chicos y las chicas jugaban
Little boys and little girls were playing
- Pocos chicos y pocas chicas jugaban
attribute is only taken as attribute of the neighbouring noun. Again, a feature. Muki987 13:14, 10 April 2009 (UTC)
- Part of the problem above is in part-of-speech tagging "Little" is being taken as as a quantifier. If you try with 'small' the result is better.
- Small boys and girls were playing
- Chicas y chicos pequeños jugaban
- Also in small case only the boys are concerned. Apertium is rather consistent here. But let's consider this as a feature, since English is also ambiguous here. Muki987 09:48, 11 April 2009 (UTC)
- The ambiguity is preserved in the sentence — this is a feature. - Francis Tyers 10:41, 11 April 2009 (UTC)
- Small boys and small girls were playing
- Los Chicos pequeños y las chicas pequeñas jugaban
- Here both of the translations are ok. - Francis Tyers 13:27, 10 April 2009 (UTC)
Expressions
What about expressions? For example "look after one's fences" this can be in the form of looking, looked, however, at present not handled at all:
- Peter looked after Martha's fences
- Peter miraba después de las vallas de Martha
The expression will be not at all recognized (Peter handled in the interest of Martha).
Is there something planned for this? Are there working examples available? 20-30% of our speech are expressions!!!! Muki987 13:38, 10 April 2009 (UTC)
Further subjects
>'házas (married- repeat all previous for this up to here, except the last 2) 1680' -- a married house? Really?
That word is a bit exception, since it has two meanings házas means married, and also a man/woman, who has a house In case if ing (shirt) inges means someone, who wears a shirt
- Ah; now I see what you mean. I thought you meant that the suffix meant married, not the word -- Jimregan 10:46, 8 April 2009 (UTC)
- Then it's a derivation, and better treated as a separate word. -- Jimregan 12:57, 7 April 2009 (UTC)
>'házacska' -- are there no lexicalised diminutives in Hungarian? I can theoretically add '-let' to any noun in English, but 'piglet' has a separate translation to most languages, and 'hamlet' is not a diminutive of 'ham'.
acska or ikó is the diminutive. It is the same thing as pig-piglet.
- I know what a diminutive is; did you understand my question? 'piglet' is a diminutive of pig, but it is a separate word in its own right, which would have its own translation -- it is lexicalised. Many (most) other diminutives are unproductive, and can be safely treated in terms of the original word. -- Jimregan 12:57, 7 April 2009 (UTC)
- Yes, there might be some words, whose diminutive form modifies the original word's meaning, however, I can't think even a single one at the moment in Hungarian. Piglet means little pig or a child pig. What do you want with these words and examples? English is very hard to translate due to tens of very different meanings of lots of words, like prime and the like. This is a very specific English problem, Hungarian or German do not have it. Are you addressing this problem? If yes, can you see any practical solution for this? Muki987 18:16, 7 April 2009 (UTC)
- You're changing the issue again. If you want a German example; 'piglet' should be translated as 'Ferkel', not 'Schweinchen'; 'Mädchen', which is a lexicalised diminutive, should not be considered a form of 'Mäd'.
- Word sense disambiguation is not a problem specific to English. -- Jimregan 10:46, 8 April 2009 (UTC)
- Not specific to English, but sharper in English, than in any other cultur language. What about your ideas to solve it? Muki987 11:54, 8 April 2009 (UTC)
- Word sense disambiguation -- Jimregan 15:52, 8 April 2009 (UTC)
- We don't currently have a good working lexical selection module, but it is one of the ideas we're hoping to get implemented through GSOC. - Francis Tyers 21:48, 8 April 2009 (UTC)
From Wikipedia:
In some cases, the diminutive suffix has become part of the basic form. These are no longer regarded as diminutive forms:
Animals
- -ka/ke: fóka (seal), róka (fox), csóka (jackdaw), pulyka (turkey), szarka (magpie)
- -cska/cske: macska (cat), kecske (goat), fecske (swallow), szöcske (grasshopper)
...which answers my question; yes, Hungarian does have lexicalised diminutives. -- Jimregan 10:50, 8 April 2009 (UTC)
- You see, you get better answers in fickipedia. You are right, this is an issue for translations, however one of the issues, that can easily be covered. Muki987 11:54, 8 April 2009 (UTC)
- Yes; it's an issue; one that you weren't considering. -- Jimregan 15:52, 8 April 2009 (UTC)
- Sure, and a lot of others also not. One after the other. Muki987 18:14, 8 April 2009 (UTC)
>Just because you can theoretically infer meaning from an analysis doesn't mean that results will translate. -- Jimregan 05:24, 7 April 2009 (UTC)
I translated so much already, that I can say: You can not say anything in any cultur language, that can not be translated into an other one.
I hope, you do not want to stress that there are untranslatable things? I would strongly disagree with that assumption, and would ask you to give me at least one example. Muki987 08:13, 7 April 2009 (UTC)
Continuation JR
- Ok, let me refine what I meant: the results won't translate in a meaningful way. There are all sorts of ways of inferring from derivational processes what a word 'means', but they tend to be useful only to linguists/translators who can then determine the best way to represent that in the target language.
- Yes, there are certain words that are not directly translatable between languages: their concepts may be conveyed in other ways, but it's an explanation, not a translation. -- Jimregan 12:57, 7 April 2009 (UTC)
- I call explanation on the target language a way to translate it. For example in German Hammelsprung means a sort of voting, when those, who say yes, exit the room using some doors, those, who say no, on some others. This can IMHO not directly be translated on any language, but must be explained; I call then the explanation translation, what it is. What do you think? Muki987 18:09, 7 April 2009 (UTC)
- 'Hammelsprung' -> 'parliamentary division', or just 'division', in context. That's the kind of translation MT should give: something as closely equivalent as possible, that fits into the same context. Your long explanation doesn't. -- Jimregan 11:02, 8 April 2009 (UTC)
- Well, that might be the case for German-English, but as far as I know, not the case for German-Hungarian. C'est la vie. Muki987 12:00, 8 April 2009 (UTC)
- Maybe. Still, it's better to use something shorter, that fits into the same general category, than to give a long winded explanation. -- Jimregan 15:55, 8 April 2009 (UTC)
- The shortest possible explanation, but it must be understandable for every reader Muki987 18:17, 8 April 2009 (UTC)
- By that, yes, you're right that if a 'close equivalent' is used, that should be as understandable as possible. On the other hand, it's perfectly acceptable to use specific terminology, which may not be understandable to everyone. -- Jimregan 08:38, 9 April 2009 (UTC)
- Yes, if one exists. I doubt, we have something in Hungarian, but I might be wrong. I can imagine "kimenõs szavazás", (voting by/at leaving) but I would not understand that without further explanations, Muki987 09:02, 9 April 2009 (UTC)
- This is rather off the topic of the discussion, this page is more to discuss methods of representing agglutinative morphology in Apertium, rather than the translation problems of agglutinative languages (which are also interesting, but better reserved for another page, or the mailing list). :) - Francis Tyers 08:21, 7 April 2009 (UTC)
- Glad to hear, that you are convinced, apertium technology is suitable for agglutinative languages. Having gone thru the English-SerboCroatian example I was not that sure. I am at the moment in the evaluation phase, and I am looking for all existing technologies. At present in my opinion google translation technology with its statistical, grammar free approach will never have the quality of a grammar oriented one, like apertium. It will for ever remain on the surface, with no real improvement perspective. However, for some situations it is very helpful. That was my first step in the direction. We can continue this subject on my discussion page, if Jimregan wants. Muki987 10:02, 7 April 2009 (UTC)
- Regarding other free grammar-focussed MT engines, you might also check out and Matxin. Open Logos has the downside of not supporting UTF-8 and not having very active development, while Matxin requires a dependency grammar to be written in Freeling format. If you want to go from English→Hungarian then this might be the answer, as they already have one written for English, but for Hungarian→English, it might take some extra development time. The Constraint grammar formalism for disambiguation and syntactic annotation might also be interesting. I'm quite happy to discuss other options and if you have any questions, please contact us on the mailing list, personally or through IRC. - Francis Tyers 10:36, 7 April 2009 (UTC)
- PS. Are you the one asking on the hunmorph list about generation ('morp visszafele')? :) - Francis Tyers 12:00, 7 April 2009 (UTC)
- Yes, I am an old language "rabbit" :-) Peter H. says, hunlex knows something similar, we are waiting for Victor, the author, he might know..... Muki987 18:04, 7 April 2009 (UTC)
Resources
Perhaps we could make a page of free resources for Hungarian ? - Francis Tyers 12:59, 9 April 2009 (UTC)
Sure, why not. As I go ahead, I'll think of the idea, and collect things. Muki987 13:31, 9 April 2009 (UTC)
Lexicon
I took a look at the lexicon you pointed me to. It looks ok, although it isn't tagged for part-of-speech. There are lots of "set phrases" (which we can extract into a translation memory) and "multiwords" (nice!) Some questions on the format:
writer szerzõ írnok író write up elõnyös színben tüntet fel feldicsér feldolgoz kidolgoz megír naprakész állapotba hoz
Is it:
<word1 in English> <trans1 in Hungarian> <trans2 in Hungarian> <word2 in English> <trans1 in Hungarian> ...
E.g. blank line, English, Hungarian ... blank line ?
- blank line is the sign for "next word" Muki987 20:11, 10 April 2009 (UTC)
Is the first translation always the most frequent one ?
- No, they are alphabeticalMuki987 20:11, 10 April 2009 (UTC)
The first step in conversion might be to take out the entries which only have one translation into a file which looks like
program ; program workweek ; munkahét workshop ; mûhely ...
Then analyse the left side with the English analyser from Apertium, the right side with the hunmorph.
^program/program<n><sg>$ ; program/NOUN ^workweek/*workweek$ ; munkahét/NOUN ^workshop/workshop<n><sg>$ ; mûhely/NOUN ^work/work<n><sg>/work<vblex><inf>/work<vblex><pres>$ ; alkotás/NOUN @ alkot/VERB[GERUND]/NOUN @ alkot/VERB[GERUND]/NOUN ...
Then extract the entries which have analyses in both analysers which agree for parts of speech.
^program/program<n><sg>$ ; program/NOUN ^workshop/workshop<n><sg>$ ; mûhely/NOUN ^work/work<n><sg>$ ; alkotás/NOUN ...
And convert that to Apertium format:
<e><p><l>program<s n="n"/></l><r>program<s n="NOUN"/></r></p></e> <e><p><l>workshop<s n="n"/></l><r>mûhely<s n="NOUN"/></r></p></e> <e><p><l>work<s n="n"/></l><r>alkotás<s n="NOUN"/></r></p></e> ...
This could largely be done automatically, but would need to be manually checked. I would focus on the most frequent open-category words, closed-category words can be done better by hand from scratch. - Francis Tyers 13:48, 10 April 2009 (UTC)
- I've emailed you a "first pass" along with some of the commands I used for creating it. If you have Apertium installed you will be able to see basic noun transfer. I still haven't figured out a way to get hunmorph to do generation though. PS. I would like to ask your permission to put the generated file in our incubator - Francis Tyers 14:47, 10 April 2009 (UTC)
- Received, thanks a million. :-) I have to install apertium on my pc, learn how to use this, and test the results. Sounds very promising!!!! Muki987 20:11, 10 April 2009 (UTC)
- Ok cool :) Can I put the dictionary I extracted in our incubator? It isn't mandatory, but it might be useful to someone at some point. - Francis Tyers 20:21, 10 April 2009 (UTC)
- Sure, you can. I think, however, it needs a lot of work before it is really usable. Muki987 21:12, 10 April 2009 (UTC)
- Thanks, and yes it definitely does, but one of my principles, is release as soon as possible :) A lot of times I've come across stuff that would be useful to me, but people don't want to release it until it is "finished" -- which sometimes doesn't happen then you lose all the work... If you rifle through the incubator you'll see a lot of junk that I've just played around with and put up there... in case someone else finds it useful some day. - Francis Tyers 21:20, 10 April 2009 (UTC)
Look after
I just spoke to another developer and he says that "vigilar" is a better translation than "cuidar". So i've added it to the SVN and the change should be live in ~12 hours.
English dictionary: <e lm="look after"><i>look</i><par n="accept__vblex"/> <p> <l><b/>after</l><r><g><b/>after</g></r></p></e> English--Spanish bilingual dictionary: <e srl="look_after"><p><l>look<g><b/>after</g><s n="vblex"/></l><r>vigilar<s n="vblex"/></r></p></e>
And now:
$ echo "Peter looked after Martha's fences" | apertium -d . en-es Peter vigiló las vallas de Martha
Much better :) - Francis Tyers 20:33, 10 April 2009 (UTC)
- Not really. Completely missing the expression meaning, he cares for her interest. vallas is fence, and not interest.
- Please also see the examples on your page, not a single working expression :-( Muki987 21:27, 10 April 2009 (UTC)
- I don't agree. I am a native speaker of English, and from that expression I take "Peter looked after the fences of Martha". If you would like to talk by telephone, I'll email you my phone number, but I think you have quite an unusual idea of what English is. - Francis Tyers 21:52, 10 April 2009 (UTC)
- By the way, in case you had any doubt, I wouldn't expect the translator to be able to translate "alri' there ma' gizzu gleg", "here y'ar" or "this int nowt" either. - Francis Tyers 22:12, 10 April 2009 (UTC)
- I also took the opportunity to ask a speaker of American English and he suggested that you might have wanted to say "mend fences" ? Which means "to resolve past conflicts, or put differences aside" -- personally I've never heard of that one, but perhaps you got mixed up? - Francis Tyers 07:53, 11 April 2009 (UTC)
- Ahh, so what I think, is a usual English expression (look after one's fences == take care of one's interest) is not a common expression at all? I must rely on dictionaries I find here and there, like Jim pertaining Hungarian grammar. So now I changed my picture, thank for your input, as native English speaker. I use skype, and we could talk over skype at any time, however, writing is probably more effective. Muki987 09:56, 11 April 2009 (UTC)
- No, it definitely isn't a common expression! :) Did you ever see the Monty Python sketch "Hungarian phrasebook (youtube)"? I don't use Skype as there is unfortunately not a free client for it. - Francis Tyers 10:13, 11 April 2009 (UTC)
- Good to know (the expression), thanks!
- That sketch is not funny for me at all.
- Skype is completely free, very-very easy to use. Telephon abroad costs always a lot, no matter how you try. Muki987 10:24, 11 April 2009 (UTC)
- No problem.
- Well, humour doesn't tend to translate well. ;)
- When I use free in terms of software, I mean free as in free software (yay ambiguity!). - Francis Tyers 10:32, 11 April 2009 (UTC)
- For me, as a poor person, free means, if I have to pay for it or not. I have not much time or interest in moralistic philosophying. Muki987 10:58, 11 April 2009 (UTC)
- If you were a truly poor person, you wouldn't be watching Monty Python videos on the internet, nor preoccupying yourself with the vagaries of English expressions or Basque linguistics. So, best to leave the high horse where it belongs, in the stable. :) - Francis Tyers 11:27, 11 April 2009 (UTC)
- No, I have never watched a Monthy Python video on Internet. I am a practicer, and it is harder to understand spoken text then to read it. I read it and did not really understand it (maybe therefore considered it as primitive hostility). Occupying myself with English expressions or Basque linguistic or translating texts is a typical activity of poor people. Nobody pays a cent today for linguistics or grammar or word collections or translations. However, I am so poor, that there are obviously people, who even envy me for being a poor person. Muki987 11:43, 11 April 2009 (UTC)
- I don't think you need to justify anything in the face of Francis' bourgeois ignorance of the vast difference between absolute and relative poverty. -- Jimregan 13:35, 11 April 2009 (UTC)
- I have an idea that might prevent you spending your time on something that isn't what you want or need. Could you send me ten sentences in Hungarian of varying levels of complexity, with linguist's glosses in English and translation in English. e.g.
- 1. John lát egy almát
- ~~ John see+VERB an+ART apple+ACC
- ~~ John sees an apple
- This would be an "easy" sentence to translate, could you send 10 examples of others from "easy" to hard? I'll then look at them and let you know which ones will be able to be translated with Apertium, which with Matxin, and which will not be able to be translated (with machine translation). - Francis Tyers 12:20, 11 April 2009 (UTC)
Examples
I have put some addresses onto my page.
- Yes, I saw :) Incidentally, our lead developer has a print out of one of the descriptions of the MetaMorfo parser on his desk. We're planning to do something similar one of these days... - Francis Tyers 21:26, 11 April 2009 (UTC)
Glosses
- John and Martha's apple and pear were sweet
- John és Martha almája és körtéje édesek voltak
- John and Martha apple-his and pear-his sweet-s be+past
- A síugrók a hegy tetején voltak
- The ski-jumpers the hill top-its be+past
- He travelled in a nice coach
- Jó kocsiban utazott
- good coach-in travel+past
Interesting here that the verb comes final. How would you say, for example "John saw Martha" and "John saw Martha through his telescope"? - Francis Tyers 19:17, 12 April 2009 (UTC)
1. example genitiv and grouping with "and"
Ok, let's start:
John's and Martha's apple and pear were sweet- John and Martha's apple and pear were sweet
- John y Martha la manzana y la pera eran dulces - Apertium
- John y la manzana de Martha y la pera eran agradables - promt
- John és Martha almája és körtéje édesek voltak - webforditas - John and Martha apple-his and pear-his sweet-s were.
- errors:
- Ignores the 's after John.- apertium
- Ignores that both apple and pear belong to John and Martha - apertium & prompt
- "John's and Martha's apple and pear were sweet" I would qualify as ungrammatical.
- "John and Martha's apple and pear were sweet". The 's possessive in English is clitic and applies to the whole phrase.
- Not that it makes a real difference to the translation quality. Co-ordinated noun phrases require extra rules that we haven't written in English→Spanish. This is not an issue of the power of the engine, but rather the number of rules. "Martha's apple was sweet" → "La manzana de Martha era dulce". The principle is the same. Incidentally, FreeLing (and therefore Matxin) also make a mess of parsing this).
- I'm calculating how frequent this construction is in the Europarl corpus, and came up with this example:
- The resources and capabilities of this country's agriculture and industry
- Los recursos y capacidades de la agricultura de este país e industria
- Los recursos y capacidades de la agricultura de este país e industria - promt
- Ennek az országnak a mezőgazdaságának és iparának az erőforrásai és képességei - webforditas
- It is still broken, but less so, again. The fixed length pattern for co-ordinated phrases after a genitive. The rule which fires above matches the pattern "DET NOM1 GENITIU NOM2", where we would need "DET NOM1 GENITIU NOM2 CC NOM3". - Francis Tyers 21:06, 11 April 2009 (UTC)
- This pattern appears in approximately 0.4% of the two million sentences in the Europarl corpus. - Francis Tyers 22:02, 11 April 2009 (UTC)
- John and Martha's apple and pear were sweet
- John y la manzana de Martha y la pera eran agradables - promt
- John y Martha la manzana y la pera eran dulces - apertium
- John és Martha almája és körtéje édesek voltak
errors:
- prompt ignores that John is also owner, ignores that apple and pear are together
- apertium ignores both as owners
2 . example word combination
- The ski jumpers were on the top of the hill
Los jerseys de esquí eran en la parte superior del cerroLos saltadores de esquí eran en la parte superior del cerro - apertium- Los saltadores de esquí estaban en la cumbre de la colina - promt
- A síugrók a domb tetõjén voltak webforditas, should be: A síugrók a hegy tetején voltak, The ski-jumpers the hill top-its were.
- errors:
Apertium does not know "ski jumpers" combination.- webforditas generates false form tetõjén ->(should be)-> tetején
- Here the Prompt output is much better. Adding the "ski jumper" multiword is easy, but that doesn't fix the problem with "en la parte superior", we'd have to look at the dictionary for that one. Again, this would be an incremental improvement, nothing "insurmountable". - Francis Tyers 21:33, 11 April 2009 (UTC)
You teached apertium? web testing shows: (http://xixona.dlsi.ua.es/apertium-www/index.php?id=translatetext):
- Los jerseys de esquí eran en la parte superior del cerro Muki987 21:41, 11 April 2009 (UTC)
- Let me update that (it updates every ~12 hours), but now I can do it manually. - Francis Tyers 21:46, 11 April 2009 (UTC)
3. example single word
- He travelled in a nice coach
- Viajó en un entrenador guapo - apertium
- Él viajó en un entrenador agradable - promt
- Egy jó edzõben utazott - webforditas - A good in-coach travelled should be: Jó kocsiban utazott: good coach-in travelled.
- Errors:
none recognizes, that coach is not only trainer, but also a coach, where I can sit. This shows, word selection (3.4 in docs) does not work reliable. Apertium also misses el. Besides that Egy is unnecessary and ugly in webforditas.Muki987 21:37, 11 April 2009 (UTC)
- Word selection is not turned on, because the current module we have does not work. I mentioned this a couple of days ago (see above in the talk page). - Francis Tyers
- I see. I will not test word selection. It is one of the most critical features generally, besides expression selection (multiword selection) Muki987 21:42, 11 April 2009 (UTC)
- Yes, when dealing with distant languages it is one of the most important features. This is why we have it as #1 on our projects list for Google Summer of Code :) - Francis Tyers 21:45, 11 April 2009 (UTC)
- Strongly agree. I would say, word selection is prio 1, expression selection immediately after word selection, and after these comes the rest. The mechanisms in apertium look good and well configurable. Muki987 21:53, 11 April 2009 (UTC)
- My main complaint after word selection (when working on unrelated languages) is that we can't (so far) do recursive pattern matching. But we're working on it... - Francis Tyers 22:01, 11 April 2009 (UTC)
- I start now installation and set up on my machine, I do not see any insurmountable problems. We shall see after setting up, how to continue. I must understand the mechanisms and handling first. If you need tests for word selection or expressions, just let me know, I am glad to help at testing. Muki987 21:53, 11 April 2009 (UTC)
- Ok, thanks. I have been working on some tricks for limited rule-based lexical selection, you can find them in category:Lexical selection. These are just proofs of concept though. Working with unrelated languages is quite new for us (yes, even after 3-4 years!), most of our work goes on related languages. PS. I would be interested in seeing glosses for the Hungarian translations above (if you have time) - Francis Tyers 22:01, 11 April 2009 (UTC)
- Done Muki987 18:48, 12 April 2009 (UTC)
- Can you check if my changes to annotations are correct? - Francis Tyers 19:17, 12 April 2009 (UTC)
Corpora
You can also add Hungarian Wikipedia to that list. We have some scripts for processing it too, see Calculating coverage. - Francis Tyers 11:20, 12 April 2009 (UTC)
Hunspell for generation
I get the same error that you posted to the list "NO DATA". The analysis works reasonably well. We would probably need to change the tagset, but that wouldn't be more than a few search/replace operations on the .aff and .dic files. "[" and "/" are reserved characters, and "<" cannot be embedded. In terms of analysis it is quite unfortunate that it doesn't do tokenise-as-you-analyse (as lttoolbox does), but it can probably be adapted to do this. - Francis Tyers 19:12, 12 April 2009 (UTC)