Features and Limitations

• This translator is "rules-based": it depends on words being where and how Turkish grammar expects them to be. This is in contrast, perhaps great contrast, to the preferred practice in machine translation, which is "example-based." This latter requires the existence of two very large corpora, as they are called, one in the source language and the other (ideally) a sentence-for-sentence mirror in the destination language. The idea is that you find in the source corpus the sentence you've been given, or something close to it, and then fish up the parallel sentence in the destination corpus, and there's your translation. If such corpora exist for Turkish and English, they are unavailable to me, and anyway, even if they were, any Turkish-translating "machine" would have to capitalize on the great straightforwardness of this language's "rules." Even if you could rig a straight example-based system, you'd want the rules too, because in Turkish - and this is what makes it so fascinating and fun - these supply so much useful and often unambiguous information about what word connects with what. In an example-based system, you'd need that grammatical facility anyway, because not every "example" is or can be available in the Turkish corpus.
  
  My sources for "the rules" are Hugo Turkish in Three Months (Bengisu Rona, 1989) and Turkish Grammar (G.L.Lewis, 1988). But my understanding and implementation of the rules are entirely my own, as are all misunderstandings and malfeasances. Don't blame these books: they are great, and you should read them in their entirety. Turkish whether or not you speak it, like baseball or football whether or not you play it, writes well.
• The other chief shortcoming of this translator, the one which really keeps it from everyday use as such, is the deficiency of its lexicons. The glossary shows all the words I have troubled to make available, and that's not very many. This'll improve, but it'll take time: I'll have to scan an entire dictionary and reformat every single entry to be readable by my own program.
• Since well below 100% of all Turkish words are represented in these lexicons, I haven't done the next obvious thing, which is to feed it stuff straight from the Internet and see when it chokes. I can, however, guarantee that all the sample sentences provided will return intelligible translations.
• Speaking of lexicons, I have provided ones for irregular English preterits and participles, that being pretty easy to arrange. You'll never see "goed" or "comed."
• Other irregularities have not been dealt with. You may see "most good." And verb-noun agreement is not guaranteed. Expect to see a lot of "is/are" and "he/she/it/they" waffling.
• You may type in your own Turkish sentences, or contribute, by means of typing, your own supplements to the samples. If what you type is in the glossary and is grammatically correct, I THINK it'll work. Who knows what you'll do, though? I have endeavored to, in programmer parlance, handle all exceptions, by which I mean the program ought not to crash; but crash or not, you may not get a translation.
• Something you should be able to do is type using all-English characters. There is a subroutine to try alternate spellings. Thus if you typed agac when you meant ağaç, the machine should figure out you meant "tree," as there is no other Turkish word within the possible permutations. But if you type sinir ("nerve") when you meant sınır ("border"), you are out of luck: the machine will leap on the first translatable spelling it finds.
• Which brings me to another feature of this and probably most machine translators: no inter-sentence assessment of context is attempted, and very little intra-sentence context assessment is done. The latter is pretty much confined to conjunctions like hem...hem ("both...and") and particles like kadar (as in xAnkara'ya vs. xAnkara'ya kadar vs. bu kadar, "to Ankara" vs. "as far as Ankara" vs. "this much").
  
  Further on the pointedly ignored subject of context, I might mention the verb oturmak, which can mean both "to sit (on)" and "to live (in)." Well, which? If you say xSnoopy evinde oturuyor, which do you mean? The machine can't know. And the only way it can even guess is if the programmer has given it lists of All The Things You Can Sit On and All The Things You Can Live In, and the program reviews both lists. These oceanic checks for what I think philosophers call categorical errors: even the most mentally underprovided biped does them all the time, while few if any computers do them at all.
• I realize just now that I've made no provision at all for interjections. The reason I realize it just now is, Turkish has quite a few which are multiword - or at least there are quite a few Turkish expressions, like afiyet olsun, "Bon appetit," which I consider interjections. And then there are the expressions which are single-word, like buyurun, "What'll it be?", whose part of speech is strictly not "interjection" but are best considered as such. Whatever you consider 'em, they ain't here. My own horseback vision is that really effective translation software would assume all sentences to be interjections, and only after certain disproof of this undertake rules-based processing.
• The program accepts just one sentence at a time, though sentences within sentences are dealt with, by which I mean quotations. IF they are enclosed within double quotes (""), that is. Observe that a single quote or inverted comma has a specific grammatical meaning in Turkish: xAnkara'ya, "to Ankara" vs. odaya, "to the room."
• The only other punctuation considered significant for translation purposes is the comma. In written Turkish its main job is to indicate the subject of a sentence has at last been got out, and/or that the word just before the comma doesn't modify one after the comma. That is how it is taken here.
• Something that doesn't have any meaning at all in Turkish is that "x" business. It is there to assist the machine, to clue it that a potentially untranslatable proper noun is on the way. The glossary has only a limited roster of the obvious ones.
• Vowel harmony matters to most speakers Turkic, but it mostly doesn't matter to this program. If you say silahlersiz when you meant (and should have said) silahlarsız, the machine should take it in stride. In point of fact, with this particular word, I DON'T know what you should have said: silah being a loan word from a language with no care for vowel harmony, it is possible that the first usage is acceptable and even correct.
• There is in the code, but not at this moment accessible, a limited ability to assess what part of speech an untranslated word is. In Turkish, nouns and verbs almost always take very distinctive suffixes, and if the machine spots those, it ought to be able to identify and handle them as nouns and verbs, without ever determining what they really mean. The feature is being withheld here because I haven't fully tested it, and maybe can't. The number of words that are Turkish is fairly large, but the number of words that aren't is infinite.
• I experimented briefly with an interactive feature, in which the program would pause when it came to an ambiguity the human user could help with. You'd tell it what you thought was the right idea or meaning or sense, and it would then carry on. As languages go, Turkish seems to present relatively few logical forks. But I dropped the idea fast. Accommodating this stuff fattened the code fast, and I doubted any likely user wants it. If you already know the right idea or meaning or sense, you probably already know Turkish.
• Just guessing, but I think the most unreliable feature of the code is its treatment of what Prof. Lewis calls izafet. A noun is said to be "in izafet with" another noun if it can be construed as the latter's possessor. E.g., "John's room," or the two halves of "bedroom." Those are simple. What's not simple in Turkish is that izafet pairs can be concatenated (i.e., the latter half of one pair is the front half of another pair), nested (one pair is inside another pair), or absent the front half (which could have been established in some preceding sentence and therefore did not, in the speaker's opinion, need to be repeated). There is also the matter of indefinite izafet, in which the front half does not carry the readily recognizable genitive ending. (Like in "bedroom" - we don't say, nor would a Turk say, "bed's room".) People can figure this out; machines, this machine at least, can't, and with some reluctance I have eliminated code meant to infer indefinite izafet. The only exceptions are çay bahçesi, "tea garden," and kuş gribi, "bird flu," which are not treated here in any izafet-related subroutine but rather in a very specialized word-pair subroutine.
  
  Izafet gets even more hairy when it comes to verbs, which can themselves be possessed by nouns. In Turkish, "You Are Here" on a city map appears as Bulunduğunuz yer, which literally means "the place where you find yourself." In Turkish, "you" OWNS "find yourself"; and as if that weren't strange enough, the result collapses to an adjective modifying the noun "place." These are called (in English, at least by the abovementioned grammarians) object participles, and I have labored considerably to get them translated because Turkish expression depends so heavily on them. The example sentences given will, as I say, come out OK. But beware otherwise.
• A comparably dicey subroutine is that which handles conjunctions. The approach is to see what's immediately west of a conjunction, assume that that is what's being conjoined, then rove east in search of a similarly situated word. For adjectives and adverbs, this is hardly fraught, but verbs and nouns present hazards, as they themselves may be but minor elements in what might be called (in view of the preceding discussion of izafet) supernouns or superadjectives.
• Just a few of the things that I just haven't got around to programming AT ALL:
  • Imperatives
  • The subjunctive...as in afiyet olsun!
  • The adjectival ending -mtrak
• Other quirks, which are at least reliable, this being a computer that's doing all the work:
  • Adverbs are made to go in front of verbs. Why? Seemed like a good idea at the time. At least I stopped the machine before it split infinitives.
  • The Turkish equivalents of "there is" or "there are" aren't quite so: you will see disagreement in number.
• If you really want to learn Turkish, you won't be spending much time on this website: you'll be in Turkey. Translation software knows where words go: it doesn't know what words do.