Mobile menu

Pages in topic:   [1 2] >
Machine Translation - A sentence to put technology to a test over time?
Thread poster: Tomás Cano Binder, CT

Tomás Cano Binder, CT  Identity Verified
Spain
Local time: 13:25
Member (2005)
English to Spanish
+ ...
Feb 13, 2009

Dear Colleagues,
As many of you, I like to put MT to a test every now and then. I do that for a number of reasons, the main two being a) curiosity and b) being able to batten the hatches if and when the technology gets dangerously ripe.

When asked for a sentence to translate, I write what I can think of at that moment, copy/paste sentences from the news or use little pieces of literature. However, my tests are far from being scientific in the sense that I cannot grasp the progress made by MT from one venue to the next. I have no fixed reference. So I was thinking of building a little list of sentences in several languages which could help us all evaluate MT options as they get available.

I am thinking of a single sentence that:
1. Is not terminologically challenging (MT will be able to toggle-in correct terminology by context at some stage).

2. Can be univocally, easily understood by any native speaker of the language. Also a sentence that is not too long.

3. Does not rely on idioms (multilingual idiom databases for MT should not be too difficult to build in the long run).

4. Is hard to grasp for a machine because of emotional, cultural, dimensional (times and places) or human-relational implications.

Can you imagine such a sentence in your language? What do you think it could be?


Direct link Reply with quote
 
kimjasper  Identity Verified
Denmark
Local time: 13:25
Member (2006)
English to Danish
+ ...
Try this Feb 13, 2009

The spirit is willing but the flesh is weak

Matthew 26:41

The old joke says that somebody suggested this phrase when the first English-Russian translation machine was demonstrated. The result was all about excellent vodka and rotten meat...


Direct link Reply with quote
 
Kevin Fulton
United States
Local time: 07:25
German to English
Time flies ... Feb 13, 2009

Time flies like an arrow; fruit flies like a banana.

Direct link Reply with quote
 

Tomás Cano Binder, CT  Identity Verified
Spain
Local time: 13:25
Member (2005)
English to Spanish
+ ...
TOPIC STARTER
Thank you, but... Feb 13, 2009

kimjasper wrote:
The spirit is willing but the flesh is weak


I was thinking more in something in your mother tongue, i.e. Danish in your case. Can you think of something that is plain to understand for a person, but plain puzzling for a machine?

(BTW: NiceTranslator.com translates "The spirit is willing but the flesh is weak" perfectly into Spanish... exactly as in the Spanish translations of the Bible... Does it use a glossary or translation memory?)


Direct link Reply with quote
 

Samuel Murray  Identity Verified
Netherlands
Local time: 13:25
Member (2006)
English to Afrikaans
+ ...
Well, you can start with... Feb 13, 2009

Tomás Cano Binder, CT wrote:
So I was thinking of building a little list of sentences in several languages which could help us all evaluate MT options as they get available.


Well, if you want a machine to struggle, just throw it a bunch of homonyms. Then add some duplication of reference words.

Apart from the spirit being willing, another classic MT breaker is "I threw a brick through a window, and it broke." So what broke -- the brick or the window? In English it doesn't matter, but I reckon in some languages it would. Ideally, if you want to get down an MT system, you should take into account both source and target language.


Direct link Reply with quote
 

LegalTransform  Identity Verified
United States
Local time: 07:25
Member (2002)
Spanish to English
+ ...
Sometimes even humans disagree Feb 13, 2009

How will a machine decide to translate something when even human beings have different opinions as to how something should be translated? Will the machine simply use the translation with the greatest number of Google hits? How dangerous.

For example, the Spanish document I am working on now contains the following term:
"la Ley del Organismo Judicial".

Although seemingly simple on the surface, a quick Google search is enough to tell you that there are many, many possible "interpretations" for this term and not all of them are good, but how is a machine to know?:

Law of the Judicial Organism
Judicature Act
Judiciary Law
Charter Law
The Judicial Branch Charter Law
Law of the Judiciary
Law on the Judiciary Body
Judicial Organization Law
Law of the Judicial Power
Ley del Organismo Judicial [leave untranslated with explanation]
etc.

Next, the computer needs to explain how this particular law should be interpreted in the context of the particular matter being discussed as well as in the context of the target language, culture and legal system.

Unless, of course, the Google Translate computer will be allowed to post KudoZ questions.

[Edited at 2009-02-13 20:59 GMT]


Direct link Reply with quote
 

Tomás Cano Binder, CT  Identity Verified
Spain
Local time: 13:25
Member (2005)
English to Spanish
+ ...
TOPIC STARTER
It's not the translation: it's about understanding Feb 13, 2009

Jeff Whittaker wrote:
How will a machine decide to translate something when even human beings have different opinions as to how something should be translated? Will the machine simply use the translation with the greatest number of Google hits? How dangerous.


Yes, that's why I specifically asked for sentences with no terminology difficulty and no idioms. Just something that, when said to you, would describe the situation perfectly as you understand the implications of what is going on. Can you think of something in that sense?


Direct link Reply with quote
 

Neil Coffey  Identity Verified
United Kingdom
Local time: 12:25
French to English
+ ...
Also look at published analyses of MT systems Feb 13, 2009

I think this could be good for fun/illustration purposes. I'd also say that the strengths and weaknesses of MT systems generally aren't too much of a mystery: with any proposed system (or proposed methodology at least), there are generally corresponding publications that describe the algorithms involved and give a measure of their performance.

Various mainstream systems including Google Translate essentially work by training a machine learning algorithm on existing bilingual corpora. To cut a hugely long story short, they produce a translation by finding one that is "statistically likely" given the body of example translations that they were trained against. (Very simplistically: "you must come" might be translated as "vous devez venir" not "vous pouvez attendre" because more examples of sentences with "you must" had translations with "vous devez" than (say) "pouvez", and more with "come" had translations with "venir" than "attendre"; similarly, "vous devez venir" is picked instead of "vous devons venir", because statistically, far more sentences are found with the sequence "vous ...ez" than "vous ...ons".)

This means that the performance of such systems is heavily tied to the features found in the training corpora. Sometimes Google Translate will appear to be performing miracles because you'll enter a sentence with various features readily found in the training material. But it's easy to make such systems go bananas by giving them, for example, sentences whose translation relies on some degree of syntactic analysis. For example, Google will translate the following:

Which waiter did you wonder whether would serve us?

as:

Quel serveur avez-vous demander si se servir de nous?

and the following:

Daniel said that flying planes is dangerous. Bob said that flying planes are dangerous.

as:

Daniel a dit que les avions volant est dangereux. Bob dit que le vol des avions sont dangereux.

As you can see, the system essentially knows no syntax.

Tomás Cano Binder, CT wrote:

Dear Colleagues,
As many of you, I like to put MT to a test every now and then. I do that for a number of reasons, the main two being a) curiosity and b) being able to batten the hatches if and when the technology gets dangerously ripe.
o I was thinking of building a little list of sentences in several languages which could help us all evaluate MT options as they get available.
[/quote]


Direct link Reply with quote
 

Tomás Cano Binder, CT  Identity Verified
Spain
Local time: 13:25
Member (2005)
English to Spanish
+ ...
TOPIC STARTER
A proposal for Spanish Feb 13, 2009

For Spanish, I was trying the following with lots of success (no meaningful automatic translation):
"Llegar tu hermana y soltarle lo nuestro fue todo uno, pero para mí que no le importa cómo me tienes."

In English it would mean something like: "When your sister arrived, you told her immediately about us, but it appears to me that she does not care about how much I feel for you".

Several online translators did this to my sentence:
"Reach your sister and drop our rioja [sic???] was one, but to me that you do not care how I got it."

"To arrive your sister and to loosen ours to him were all one, but for me who does not matter to him how you have to me."

"Your sister to come and ours to give up him was every one, but for me that does not import for him(her) how you have me."

"Get your sister and drop him our was around one, but for me that don't care how I have."

"Come tu sibling and loosen him our it was all one , but as for me which nay her levy how me has."

So far, so good!




[Edited at 2009-02-13 21:33 GMT]


Direct link Reply with quote
 

LegalTransform  Identity Verified
United States
Local time: 07:25
Member (2002)
Spanish to English
+ ...
Sentence Feb 13, 2009

The sentence "I never said she stole my money." could have several meanings depending on the surrounding text (or emphasis when spoken):

* "I never said she stole my money." - Someone else said it, but I didn't.
* "I NEVER said she stole my money." - I simply didn't ever say it.
* "I never SAID she stole my money." - I might have implied it in some way, but I never explicitly said it.
* "I never said SHE stole my money." - I said someone took it, I didn't say it was her.
* "I never said she STOLE my money." - I just said she probably borrowed it.
* "I never said she stole MY money." - I said she stole someone else's money.
* "I never said she stole my MONEY." - I said she stole something, but not my money.

Tomás Cano Binder, CT wrote:

Yes, that's why I specifically asked for sentences with no terminology difficulty and no idioms. Just something that, when said to you, would describe the situation perfectly as you understand the implications of what is going on. Can you think of something in that sense?


[Edited at 2009-02-13 21:35 GMT]


Direct link Reply with quote
 

Tomás Cano Binder, CT  Identity Verified
Spain
Local time: 13:25
Member (2005)
English to Spanish
+ ...
TOPIC STARTER
Yes, but... Feb 13, 2009

Jeff Whittaker wrote:
The sentence "I never said she stole my money." could have several meanings depending on the surrounding text (or emphasis when spoken):


I agree, but it would be a matter of intonation, and the words would be the same in another language, with different intonation of course, if autotranslated, IMHO of course.

I was thinking about something with relational implications that created trouble for an automatic translator. I have posted a proposal for Spanish, but made a mistake stating what autotranslators had done what and must be now be vetted...


Direct link Reply with quote
 

Sheila Wilson  Identity Verified
Spain
Local time: 12:25
Member (2007)
English
+ ...
Nothing to add to thread, but thanks, Jeff Feb 13, 2009

Jeff Whittaker wrote:
The sentence "I never said she stole my money." could have several meanings depending on the surrounding text (or emphasis when spoken)


Thanks Jeff. If you don't mind I'll hijack that one for my English lessons!


Direct link Reply with quote
 

Samuel Murray  Identity Verified
Netherlands
Local time: 13:25
Member (2006)
English to Afrikaans
+ ...
Indicating intonation Feb 13, 2009

Jeff Whittaker wrote:
* "I never said she stole my money." - Someone else said it, but I didn't.
* "I NEVER said she stole my money." - I simply didn't ever say it.
* "I never SAID she stole my money." - I might have implied it in some way, but I never explicitly said it.
* "I never said SHE stole my money." - I said someone took it, I didn't say it was her.
* "I never said she STOLE my money." - I just said she probably borrowed it.
* "I never said she stole MY money." - I said she stole someone else's money.
* "I never said she stole my MONEY." - I said she stole something, but not my money.


Afrikaans is one of those languages in which it is quite normal to indicate stress, using acutes on the vowels.

* "I never said she stole my money." - Ék het nooit gesê sy't my geld gesteel nie.
* "I NEVER said she stole my money." - Ek het nóóit gesê sy't my geld gesteel nie.
* "I never SAID she stole my money." - Ek het nooit gesé sy't my geld gesteel nie.
* "I never said SHE stole my money." - Ek het nooit gesê sý't my geld gesteel nie.
* "I never said she STOLE my money." - Ek het nooit gesê sy't my geld gestéél nie.
* "I never said she stole MY money." - Ek het nooit gesê sy't mý geld gesteel nie.
* "I never said she stole my MONEY." - Ek het nooit gesê sy't my géld gesteel nie.

And one more not included in your list:
* "I repeat, I never said she stole my MONEY." - Ek hét nooit gesê sy't my geld gesteel nie.


Direct link Reply with quote
 

José Henrique Lamensdorf  Identity Verified
Brazil
Local time: 08:25
English to Portuguese
+ ...
They fixed it later Feb 14, 2009

Tomas, I hope you'll get it from Portuguese via Spanish.

Earlier machine translation used to render...
ACME utilizes the latest technology...
in Portuguese as:
A ACME utiliza a tecnologia mais atrasada...

Now most have fixed that, but it was funny to see that on machine-translated web sites boasting their gizmos' features.



It's foul play, but I did it in EN-ES with Babelfish:
The murder weapon had been left right there in plain sight to be picked up.
became:
El arma de asesinato había sido izquierda-derecha allí en la vista llana que se cogerá.


It's worse than I would imagine a Portuñol class:
MIENTRAS (as explained by a Brazilian): "Se tu 'mientras' por la puerta, yo voy a salir por la ventana."


Direct link Reply with quote
 
Pages in topic:   [1 2] >


To report site rules violations or get help, contact a site moderator:


You can also contact site staff by submitting a support request »

Machine Translation - A sentence to put technology to a test over time?

Advanced search






Protemos translation business management system
Create your account in minutes, and start working! 3-month trial for agencies, and free for freelancers!

The system lets you keep client/vendor database, with contacts and rates, manage projects and assign jobs to vendors, issue invoices, track payments, store and manage project files, generate business reports on turnover profit per client/manager etc.

More info »
Across v6.3
Translation Toolkit and Sales Potential under One Roof

Apart from features that enable you to translate more efficiently, the new Across Translator Edition v6.3 comprises your crossMarket membership. The new online network for Across users assists you in exploring new sales potential and generating revenue.

More info »



All of ProZ.com
  • All of ProZ.com
  • Term search
  • Jobs