4

Many people i see over online auction sites and such use software translators for their benefit, but sometimes the outcome can be somewhat curious. Are these inaccuracies caused by the use of informal input language or a lacking algorithmic structure? In case of the latter: which languages are most far apart and arduous to translate directly from one to another for the software?

  • 2
    Consider the problem of statistical MT for languages with only small or no appropriate corpora available. – snailplane Jun 23 '14 at 2:46
  • 1
    As it is written, this is really a very opinion based question and probably unanswerable. What would you think about changing it to "What factors make software translation very difficult?" – curiousdannii Jul 3 '14 at 22:11
  • Google translation from English to Russian does not work totally, yet translation from Russian to English seems to more or less work. – Anixx Jul 16 '14 at 15:24
  • @robert your comment must have gone to the wrong place – curiousdannii Jul 17 '14 at 3:52
4

It depends on generally syntactic structure of two languages and -yes- some pragmatic issues. To answer your question,

  • the easiest languages to be translated into English by any automatic translation software are the ones which are nearly close to syntactic structures of English such as German, Chinese blah blah.

  • the hardest ones are the ones which are far away from the target language's (which means English) syntactic structures such as Turkish, Japanese, etc. (because they are agglutinative languages).

It depends on some pragmatic issues, though. For instance, if you use a informal text in German and translate it into English, the output will seem to be odd. So there is a possibility of translating by a software. Sometimes you even get strange outputs even though two languages have the same typology, so there is no hundred per cent translation on these softwares.

| improve this answer | |
  • I find that even Google Translate is still quite bad at German and is easily confused by the different word order such as finite verb second, nonfinite verbs at the end, and separable verb prefixes. I also find it's bad at Chinese, probably due to lack of word breaks leading to more ambiguities as one of the factors. – hippietrail Jun 19 '14 at 6:27
  • 1
    Isn't German quite different in grammar than English? I would expect Scandinavian langauges to translate much easier than German. – Midas Jul 4 '14 at 6:26
  • I do not have any clue on Scandinavian languages. In German to English example, I just thought of verbial sentences, their transformations and appearance of their predicates in those transformations. – Eray Erdin Jul 5 '14 at 17:01
  • 1
    @ErdinEray: Well from my experience I find it very easy to translate between them without any major mistakes. – Midas Jul 5 '14 at 18:58
3

"The hardest" is difficult to say. I will say that online translators often seem to have especial trouble with Japanese, as its often implicit anaphors (subjectless sentences and so forth) are extremely difficult for a machine algorithm to infer, so you often get lots of confused pronouns, mixed up genders, etc.

| improve this answer | |
  • The lack of word breaks and the very different word order no doubt also contribute to the difficulties MT have with Japanese. – hippietrail Jun 19 '14 at 14:02
  • 2
    Polysynthetic languages should also be quite challenging. I write should because I don't know of a software attempting translation from say English to a polysynthetic language. – Olivier Jun 19 '14 at 15:24
1

The most challenging (not to say, practically impossible with current development of NLP) is to translate between languages that are built around completely different concepts.

As for Indo-European languages or languages like Chinese, Georgian, Swahili, the concepts are similar. You have nouns described by verbs and adjectives, those are described by adverbs. There are pronouns. There are prepositions and other modifiers, which can be attached as separate words or as prefixes or suffixes.

To find something very different, look to polysynthetic languages, which have a large number of prefixes and suffixes, much bigger than in agglutinating languages like Hungarian. Look for example at the Navajo language. It is organized around verbs, not nouns. There are verbs that provide the functionality of adjectives. There are numerous moods and aspects.

To my knowledge, even creating NLP tools to analyze a polysynthetic language is a big challenge, compared to NLP parsers for other group of languages. I'm not sure if translating between polysynthetic languages would be easier than between polysynthetic and non-polysynthetic language.

Limited corpora for such languages is another factor, that makes that task even harder.

| improve this answer | |
0

I suppose the hardest ones are Agglutinative languages like Japanese, Korean, Turkish etc.

If you use an agglutinative language (even you are using very formal wording) and don't check resultant sentences, translators can make very big mistakes. If I take Google Translate into consideration, the problematic part is to make translation engine to understand words and their structure correctly.

Source ----understanding--->   Meaning  ----translation----->  Destination

Because of the hardness of the problem, automatic translators can't split words into its parts correctly, especially in agglutinative languages.

To solve this problem, you can use your own algorithms your own corpus. My approach is to revise source sentences. I split long sentences into small ones, changing complex source words with simple ones. If you make translator to understand the meaning, it can successfully translate it to any language. Though, these revisions make source text too simple too read. But it is preferable if you use automatic translators frequently.

| improve this answer | |
  • 2
    Why would agglutinative languages be harder than other languages? Polysynthetic languages must be hard to translate too! – curiousdannii Jul 16 '14 at 7:02
0

The difficulty arises not from the various linguistic properties of the language pair as many here suggest.

Google machine translation direction for German to Chinese pair looks like this:

German -> English -> Chinese

Where English is a bridge language, a language that is used in between German and Chinese. This is done to save resources and development time, so that instead of creating segmentors, tokenizers, parsers, shallow/full syntax rules for every language pair combination you create these linguistic resources for just one.

Here is an example. We shall try to use translate.google.com to translate a phrase "to lose a suit" from Russian to Chinese:

проиграть процесс -> 失败西装

失败西装 in Chinese is "to lose a jacket", which happens because suit is an ambiguous word in English:

"проиграть процесс -> fail in a suit -> 失败西装"

So answering your question, I'd say that the hardest language to translate from is the one which has a higher syntactic/semantic ambiguity. That's why using English as a bridge language can cause translation inconsistencies as the above.

| improve this answer | |
  • I have to vote you down because your claim that "this is not how MT works" is just wrong. Google Translate does currently use English as an interlanguage when translating between two other languages, but they could've made any other decision and may make any other decision in the future. It is true that it's done to save resources and development time. I also agree with your conclusion that higher ambiguity definitely is a big factor. – hippietrail Jul 16 '14 at 14:09
  • an overstatement of course, fixed. – tasty Jul 16 '14 at 14:19
  • I wanted to come and vote you up but there are still overstatements such as "Machine translation algorithm of German to Chinese pair looks like this". – hippietrail Jul 16 '14 at 14:22
0

It is not really any given language that is hard to translate. The problem is contextual words or phrases that mean different things (either in English or in the language being spoken). This includes colloquial phrases, cultural differences (e.g. the boot or trunk of a car). There are also domain specific differences (e.g. "International Phonetic Alphabet" in linguistics and the navy).

Consider the phrases:

  1. she took the lead in the dance;
  2. he took the lead in the box to the dog's new owner;
  3. they took the lead in the box to be analysed.

Here the phrase "took the lead in the" is the same, but 'lead' is referring to 3 different things:

  1. is referring to leadership;
  2. is referring to a dog lead or collar;
  3. is referring to the metal.

There are also phrases like "that's rubbish", "that's pants", etc. that can indicate that something is not very good. Likewise, phrases like "that's mint", "that's the bomb", "that's cool", "that's wicked" can indicate something is very good.

Also, with "wicked" it can also mean the musical based in the Wizard of Oz universe, among its other definitions. Likewise, there is a German band called Die Toten Hosen ("the dead trousers"), which Google translate leaves in German.

Interestingly, while "that's pants" translates to "das ist Hosen" in German, it translates back to "that is pants" in English, mainly because it preferentially uses pants over trousers (e.g. translating "Hosen" to English), due probably to its US English slant (pants mean underwear in England).

Software translators also do a poor job when encountering poor grammar (e.g. not capitalizing the correct words in German, leaving out punctuation, etc.).

I would suspect that Hebrew would be difficult because from what I understand it tends to leave out the vowels.

| improve this answer | |
  • "I would suspect that Hebrew would be difficult because from what I understand it tends to leave out the vowels." - You're confusing language and orthography. But it's actually worse than this. There are various kinds of Hebrew orthography depending on whether you leave vowels out, use consonants for some vowels, use diacritics to indicate vowels, or use some combination of both systems together. Arabic has some of the same problems and options but seems to have only one orthography. Of course it has a lot more variety in the language than in the orthography. – hippietrail Jul 19 '14 at 4:14

Not the answer you're looking for? Browse other questions tagged or ask your own question.