Systran has been around for ages already, it's not a new engine by any means. As far as I know it's also the same translation engine that ATLAS is build on (the translation software, not the script inserter...). Judging from my experiences with AGTH it's definitely not near the quality required to use it for anything but looking up a few words. It's better than Babelfish though.

 :banghead:

 

Then I mixed things up. Do you know which engine Atlas is based on? I'm pretty sure it's something different then.

eh, that’ll show me to trust kotoeri again... *mumble mumble*

Well, it just sucks, and that’s that. But ultimately, kana vs. kanji is a much harder thing than mere statistics. Computers aren’t so great at figuring out where one word stops and another begins in Japanese; humans can only do it so well because we’re cacophonologists by nature.

Atlas (or perhaps more properly “Atlas II”) is its own system. What you can get today is fundamentally the same thing that went to market in 1982, except now you get the benefit of a couple decades’ worth of tinkering and UI changes, as well as some big-ass dictionaries.

Atlas is rule-based, rather than statistical.

I double checked it with RikaiChan which is a dictionary plugin for Firefox, but you probably know that since you all seem pretty familiar with this stuff. It also picks up the "to release" definition from  放す and  はなす but I can see that 放 is not even in the sentence.  But the next definition is "to speak" which it picks up from 語す and  はなす and I can see those are all in the sentence but in different parts.

It's too bad words or even phrases are not seperated in writing.  That surely would make it easier although I'm guessing the sentence becomes too complex in some situations like in this example.

Hmm, now I'd like to know what are the most used kanji, period.

You overestimate SYSTRAN’s ability to tell word boundaries in Japanese, though.

It didn’t just fail at translating a single word; the bigger problem is that it failed to translate the grammar naturally. If you type “日本語、話せる？” you get “Japanese, you can speak?” and other bits of Tarzan-speak. Note how none of your examples used the proper English: “Can you speak Japanese?” SYSTRAN is so afraid to switch sentence order around in anything but where it is strictly permitted, that in this case, it merely translates the sentence and then adds a question mark at the end for か. On one hand, this has a lot to do with its original focus in assisting translation with such clients like the EU and the UN that need to pass around technical documents and legal crap in multiple languages. On the other hand, it makes it utter poo for conversation and literary works (like video games, unlike what Old Man Ebert may tell you), and SYSTRAN’s academic papers would be the first to mention this.

Let’s take first a simple sentence and then a complicated one from my current project (Ace Combat 3):

もはや一刻の猶予も許されない。
Either postponement of moment is not permitted already.

The “もはや” is “already” and gets shoved to the end. “一刻の猶予も” is translated “either postponement of moment”, which makes a little bit of sense translating literally, but is ultimately gibberish. “許されない” is translated “is not permitted” and is shoehorned in right before the end, which is technically fine, but the meaning, in context of the “も〜ない” is more like “cannot permit” when you put it in English, despite this not being grammatically equivalent.

Now let’s play dirty:

死因も……どうやら人工心臓のネットワークに接続されていた自動制御プログラムが、何らかの原因で暴走した事によるものと見られ、関係者の間でも大きな波紋を呼んでいます。
Cause of death ......Somehow, the automatic control program which is connected to the network of the artificial heart, it is seen as the thing due to driving recklessly with a some cause calls the big ripple even between the authorized personnel.

Well, this is where a rule-based system like SYSTRAN shines. Except for a little word->word boo-boos, and a couple of grammar niggles, this is actually reasonably close to how I translated it. All that cautiousness pays off if you end up with something that remotely makes sense.

Cf. Google, just for kicks:

もはや一刻の猶予も許されない。
Not allowed time to lose anymore.

Equally bad as above, but in some different ways.

死因も……どうやら人工心臓のネットワークに接続されていた自動制御プログラムが、何らかの原因で暴走した事によるものと見られ、関係者の間でも大きな波紋を呼んでいます。
Cause of death ... ... and Control Program also had an artificial heart is connected to the network apparently believed by the runaway thing for whatever reason, is causing a big stir among the people concerned.

Close, but no cigar. The part “is causing a big stir among the people concerned” is great, but “apparently believed by the runaway thing for whatever reason” is aswingandamiss; the “による” can mean “by” or “on account of”, which is basically the same thing, except for a subtlety in English in which “by” used with the passive voice indicates that something is directly responsible (and would be the subject of the sentence were the active voice used). In this case, Google Translate botched it, because the “と” is more important here, and SYSTRAN’s “it is seen as” is much closer to the mark.

---
As to はなす, there is a simpler explanation: usually, when kana only are used instead of a common kanji, it’s because they wanted a complex kanji but didn’t want to bother fishing through the IME to select it. SYSTRAN’s rule-based system, with its gazillions of language pairs, reflects this. Google’s statistics-based system does not, and just goes with what it thinks would make sense statistically.