Foreign (non English speaking) Members

[Chris Soucy]

We may yet get a live one (Romulan), in which case we can leave the Klingon translator in and watch the continuing series of Star Trek as they have at it.

I thought this was going to take a swerve into the exotic when Ervin mentioned missionary, but alas, it did not transpire.

The offers of help were a great encouragement however, thanks guys.

On a more serious note, the body, er, view count is racking up nicely, but there's not nearly the input coming in I was expecting for such a mind bogglingly possible expansion of DVinfo's capabilities.

To quote from a very well known Australian tourist ad:

"Where the bloody hell are you"

And as for setting out on this journey, well, I'll quote another well knowm OZ tourist ad:

"You'll never never know, if you never never go"

Come on guys and gals, give us some feedback/ input.


CS

[Bob Hart]

Ngai brubi learnaman wadji fillummaka. Ngai jaja fillum nyarmbali yaala.

I'll leave you guessing on this one. In the second Starwars, there were two short phrases in this language which made sense contextually so I assume the writer dug deep and wide through uncommon languages to create Huttese.

[Ervin Farkas]

Kion pri Esperanto? Estus bela mezo tero...

[How about Esperanto? Would be a nice middle ground...].

[Gareth Watkins]

The Dutch isn't the only totally gobbledygook translation... the French is just as bad... good job I had the English one to look at to see what it was all about...

Guess the perfect automatic translator has yet to be invented.

Amitiés
Gareth

[Chris Soucy]

amongst us.

Further reflection on just why this translation stuff is failing so badly has led me to the inescapable conclusion that a major part of the problem is the sheer plasticity of English itself.

[There, case in point. I've probably broken a half dozen "rules of English" in that sentence alone, no wonder the translators are falling off their perch. I doubt that any of the English as a first language readers even batted an eyelid].

Which led me into a bit of a minefield wondering whether other languages are more "structured" and thus less likely to be mangled by the people using them.

So, thought I, what happens if you make up a sentence, no longer than, say, 15 words maximum, which is totally incapable of being misconstrued (in English), obeys all the rules, has no spelling mistakes, contains no ambiguity whatsoever, has minimal punctuation and delivers the message loud and clear.

Now, my logic says that if that sentence is run through the translator, the result should be just as unequivacal as the original (ah, but is it?).

Let us assume the previous assumption is correct.

If it is, then running the translated sentence back through the translator should return you to the original English version in perfect health.

Take this sentence:

The quick brown fox jumped over the lazy dog.

All of you that speak foreign, type it into the translator (go to Google, it's on the tool bar, top left) and translate it into your favourite "other language".

How does it read?

Now, copy the result, delete the original and paste the result into the text box. Reverse the language from/ to selection.

Hey presto, in all the languages I've tried, I get the original back.

Now try this one:

Mary had a little lamb, it's fleece was white as snow.

I won't beat about the bush, it fails.

Why?

Because of that comma.

It should read:

Mary had a little lamb. It's fleece was white as snow.

Most of us that use English (as a first language) have forgotten so many basic rules we don't even stop to think about how badly we're mangling it, because we can navigate quite happily without them.

So, where am I going with this?

I'd like anyone who has some spare time and a bit of curiosity to try out their own sentences (any and every language you like) and test them on the translate/ re - translate torture rack and see if my premise holds true if the rules are obeyed.

So, what's the bottom line here?

Quite simply, if my theory is correct, running the tranlate/ re - translate routine as you're typing, so that each sentence is displayed back to you in the source language, would very quickly flag where the rules were being broken, effectively re - teaching you, er, me, how to write according to the rules.

[See, there, I've done it again - 3 1/3 lines (in the post box, maybe 2 on the page) and not a full stop to be seen and probably utter giberish on it's return from the translator, but, then, maybe not.]

To put it another way, as it looks like we'll never teach a machine based translator to accurately decode a moveable feast such as coloquial English, maybe the translator can teach us to write within the rules and give it a sporting chance.

Thoughts?


CS

[Trond Saetre]

I am just glad I do not have to read a Norwegian translation of this site.
I have read, or tried to read, some Norwegian technical books about both computers, electronics, video...
And honestly, it is much easier to understand the English tech words than translated words.

And as been said already, I have yet to find an online translator which can do a fairly good job with more than only basic sentences.

[Chris Soucy]

Put my money where my mouth is and gave it a try.

http://www.dvinfo.net/forum/open-dv-...ml#post1468380

Check out post #11.

If ever there was a time to get this thing working, this thread is it.

Took the basis from a previous post further up the page, pasted it into Word, tweaked it a tad then stuffed it into Google and translated it to Swedish.

Translated it back to English where, much to my amazement, it had actually improved the wording of the first line significantly.

Changed the first line to the suggested, added a bottom line and did it again through Google.

It came back out of the wash looking pretty good so decided to fire it off to Sweden.

What the reaction there will be is anybody's guess.

Not a fandango you'd want to do too often, but if a system could be arranged whereby a spell check, grammer check and the translate/ re - translate worked concurrently as you typed, could be very effective indeed.

I have yet to figure out just why the differences I see between the original and the subsequent re - translate back to English are as they are, that makes no sense as far as a rule base goes, but something that could easily be adapted to in a short space of time.

The down side of it all is that with all three running concurrent with one's typing, it'd be a bit like having an entire car full of back seat drivers.

The up side is it'd be so painfull that in pretty short order one would learn to obey the rules and they'd shut up.

In case anyone is having a problem with this concurrent translate/ re - translate concept, imagine the white box you see when typing a post.

Now imagine another seperate white box under/ over it which is being filled with English text from the tranlate/ re - translate routine for each completed sentence or usable part thereof.

Instant feedback of translator bloopers.

The theory being that if your text can be translated from English to (pick a language) and back to English with some semblance of sanity, it's fit to go out to all the other possible languages that might be on the system.

Hey presto, a multi lingual DVinfo that teaches you how to work it.

Think it might be good to get some feedback from Chris Hurd before going too much further down the track. Anyone know if he's gone/ going to CES?


CS

[John McCully]

Chris Soucy, I can not resist joining, momentarily, this conversation. Not because I can offer any constructive advice regarding your goal with this thread but because I found your wanderings into Literary Theory interesting. You got my attention; pondering the notion of translation and translatability all within the context of cultural and phonetic differences, fairy stories, a little lamb (glad you added the lamb; very New Zealand) not to mention the miss-connects that occur between like peoples because of traditional beliefs, dogmas, and other non-touchable ways of interpreting. And what a difference a full stop makes, not to mention that dastardly comma, that often inserts a pause where one might rather have moved right along.

I also detected in your posts in this thread a touch of the same predisposition towards critical inquiry that leads you down the path of tripod deconstruction.

Good for you.

I discovered a downloadable series of lectures at Yale University. Digression: It is totally wonderful that the USA in the form of Yale give away lectures on many subjects. It doesn’t come any better. And speaking as a Kiwi Canadian I say to all those who sneer at the States, seems there’s no letup, to them I say go online to Yale University and check what that American institution is doing, at no cost to the downloader. I suggest it is incredibly generous, thoughtful, living breathing Human Rights on a grand scale. The US takes a lot of criticism and sometimes I like to point out the other side of the coin, especially when it benefits me, as in this instance. I’m doing ‘An Introduction to Literary Theory – Paul H Fry’. I’ve already downloaded 4 lectures and the text book for the series will be here in about a week. So after I get through that, 26 lectures and some heavy reading, I might have more for you on the business of translation. In the meantime I might take your words ‘have forgotten so many basic rules we don't even stop to think about how badly we're mangling it’, (the English language you are talking about) and ask Professor Fry how he feels about that notion. After I do this course I’ll report back with perhaps a sensible suggestion or two about the task of the translator, sliding signification and the mangling business.

I’d better quit this line of reply before Chris Hurd says ‘look boys, if you want to have that kind of conversation could you go somewhere else, please’, whereupon I’d quickly reach for my EX1 and let him have it, handheld, 24p, (the film look). If that didn’t work you could come to my rescue and with one fluid pan of your Vinten carbon-fibre quick-lock legs you would take the wind right out of his sails, with a smile. No worries.

Cheers

John

[Chris Soucy]

What you smoking up there? (ROFLH).

Can you send a small parcel down here? You've got the address.

I'll most defiantely have what he's having!

I think I'm gonna need all the help I can get with this one.

Thanks for one of the most erudite responses to a thread I've ever read, good on you mate (us Kiwi Canadians have to stick together).

Don't think you have to worry too much about CH, he's most probably watching, waiting to see where it goes, if anywhere, and then he'll chime in, no hurry.

Hey, he's threatened to visit us in NZ on hols one year, he won't blow that on a whim!

Enjoy your Yale lectures.

Regards,


CS

[Trond Saetre]

Originally Posted by Chris Soucy

Now try this one:

Mary had a little lamb, it's fleece was white as snow.

I won't beat about the bush, it fails.

Why?

Because of that comma.

It should read:

Mary had a little lamb. It's fleece was white as snow.

When translating these to/from Norwegian, the words "It's fleece" was not translated from English, using Google translator.

[Colin McDonald]

I try to switch off from teacher mode out of hours but I may be able to clarify this one:

Mary had a little lamb. It's fleece was white as snow.

When translating these to/from Norwegian, the words "It's fleece" was not translated from English, using Google translator.

"It's" is an abbreviation for "it is" NOT a possessive.

So the second line of the nursery rhyme should be "Its fleece was white as snow."

/pedant mode

[Ervin Farkas]

Correct. I tried going English to German and back with the corrected sentence, and "Mary hatte ein kleines Lamm. Sein Fell war weiß wie Schnee" was translated back into the exact same English phrase.

Bingo,

[Chris Soucy]

Well spotted.

You know, I looked at that about a dozen times whilst I was playing around and KNEW something wasn't right, even though it was staring me in the face.

Hmm, spelling, grammer and punctuation - perfection required.

Somebody please tell me that other languages aren't this easy to mangle!

On another note, it has occured to me that this process can be made slightly easier if the translator itself is "trained" to reckognise slightly more diverse text than it currently does.

If anybody read the article on Machine Translation I linked to, they will remember that Google got their system much improved by feeding it 200 BILLION words from UN documents.

All very well, but I doubt we'll see writing like that here any time soon.

However, how many words are there in the DVinfo post backlog?

That 1 million plus posts must contain a heck of a lot of coloquial English.

Maybe...............

But, then again, maybe not!

The problem, of course, is that the 200 billion words would have been taken from documents translated by many highly qualified human beings at a cost I don't even want to think about.

The DVinfo archive is English and only English.

Back to the drawing board.


CS

[Chris Soucy]

Having reached something of a hiatus on the subject, a little bit of conjecture to be going on with.

Disambiguation.

Great word, and probably one of the toughest nuts of all to crack, especially if the translator has no access to the author, as is the case with standard document based translations. Anyone who read this article: Machine translation - Wikipedia, the free encyclopedia in one of my earlier posts may remember the translation pyramid it contained, which shows source text being analysed into an interlingua before being translated into the target language.

Now, for our purposes, the authors (you and I) are not only available but actually in the process of creating the text to be translated, it is thus capable of being altered in its entirety.

Now, assuming we have a spell checker, grammar corrector and punctuation policeman from hell, all we’re left with is the dreaded disambiguation.

As was demonstrated previously, if a sentence can be produced with perfect spelling etc and contains no ambiguities, the statistical translator in Google is pretty well spot on as shown by translating the sentence to another language and back again to receive the same text as transmitted.

So, how do we solve the ambiguities conundrum? If you take a sentence and pass it through the analysis engine towards the interlingua stage, an ambiguity must create 2 or more possible results. Take the word pair “going out” for example. Is this used in the context of “the fire is going out” OR “We’re going out tonight”, describing two completely different things (fire and socialising), which to a machine make no sense at all?

However, let us assume a sentence is designed which follows all the rules and contains just one ambiguity. Now, if the text is passed through the translator to the interlingua stage and all of the possible generated options are fired back into English and compared to the original, one should be correct and the others should not, easily measured.

So, if, as we are typing, the background software continually took the sentence, passed it through the analysis phase, passed it back into English and compared all the possible results with the original text, the results should be capable of generating a statistical ranking based on the original text, 100% being the goal but I’d settle for 95%. As the author is right there, a major blooper caused by nested ambiguities, say, which gave a stats figure of less than the chosen 95%, can be corrected on the fly, tested and either changed again or accepted.

This sounds all very long winded and is certainly not something you’d do manually, but heck, you’ve got that quad core and 8 Gig connected by megabit broadband to Chris’s server farm and it’s doing nothing but waiting for you to hit a key twice a second at best.

Note that I have made no mention of Google. I envisage this software app to be on Chris’s servers and freely downloadable. The actual translation to other languages happens after your post is done and dusted on said server(s).

Now, you may have noticed I didn’t suggest passing the text through the translator to another language and back? The reason is simple, I’m certain there are many, many English words that simply cannot map to a word or phrase in other languages. If it fails there it must fail hopelessly going back again, so it’s actually introduced an error needlessly. Quite how the Google translator handles this situation I’d be interested to know.

There is very likely a good reason why the system doesn’t or couldn’t work this way, but it does seem like a logical presumption to make given the above.

Just a little something to keep you thinking.

CS

[Marty Welk]

exactally ^

If you write simply, it can be translated.
If the translator was phrase based, the phrase would still have to be in the database.

With a constant operating live cross translation, feedback to the original language, a writer could create sentances that conformed to the ability to translate, by thier choice of words and phrases.

What would it be like to write about video, so it reads like a childrens book?
The fur on the sheep was white. Snow is white. The fur and the snow were similar.

I have a metal object. The object has 3 legs. There is an item on the top of the object. The item on top will move up and down. The item on top will move left and right. The item is a Head. The item is not your head. the item is not my head. I want to make movements that are smooth or soft. . .