Neural machine translation offers significant advances with remaining challenges
Thread poster: Hans Lenting
Hans Lenting
Hans Lenting
Netherlands
Member (2006)
German to Dutch
Mar 12, 2018

NMT does not learn languages in the same way humans do, even though some breathless reporting has made this claim. Instead, it relies on statistical correlations, much like PbSMT does. The difference is that NMT can make much more complicated inferences from that data and is very good at determining correlations of correlations. Where a PbSMT system might observe that English frog tends to translate into German as Frosch, a neural system could note that if the text mentions words to do with railways, Herzstück is a much more likely translation, even if this translation occurs only a few times in a training corpus.

Older systems look at n-grams, strings of a certain number of words. For example, if a system works with 6-grams, it considers chunks of up to six words. This approach works fine for linguistic structures that are compact, but has trouble with "long-distance dependencies," such as German verb phrases, which may contain entire clauses in between parts of a verb phrase. By contrast, NMT systems look at whole sentences in their entirety, and researchers are now pushing them to work on entire paragraphs or even longer chunks of text. This shift allows them to be more sensitive to context and handle complex grammatical structures more effectively.

NMT looks at individual characters, while phrase-based approaches look at words. This difference makes neural systems particularly good at working with morphologically rich languages, such as German or Hungarian. For example, a PbSMT system would not – without additional language technology – recognize that both speichern and gespeichert are forms of the same verb. By contrast, NMT can work with patterns of characters to predict word forms it may not have previously seen.

Neural systems can extrapolate across multiple languages to fill in gaps in training data. This capability called "zero-shot translation" allows NMT engines to translate language pairs for which they have no data or to fill in gaps in training data from other language pairs. For example, if a NMT engine has English<>Greek and English<>Finnish training data, but no GreekFinnish, it can use the information from its existing language pairs to translate that pair. Although the results will not be as good as for pairs where it has data, this can make the difference between having some translation and no translation at all.


Source: http://www.tcworld.info/e-magazine/translation-and-localization/article/neural-machine-translation-offers-significant-advances-with-remaining-challenges/

[Edited at 2018-03-12 07:45 GMT]


 
Egidijus Slepetys
Egidijus Slepetys  Identity Verified
Local time: 22:38
German to Lithuanian
"contactor" - "Abdullah". Mar 16, 2018

The term "Neural machine translation" is a joke. It's a scam to attract owners of translation agencies, who didn't translate a word in their lives.

 
Hans Lenting
Hans Lenting
Netherlands
Member (2006)
German to Dutch
TOPIC STARTER
Neural machine translation is there for us Mar 16, 2018

Egidijus Slepetys wrote:

The term "Neural machine translation" is a joke. It's a scam to attract owners of translation agencies, who didn't translate a word in their lives.


Neural machine translation is there to assist us, translators, while working on individual segments in a CAT tool. It's not suited to translate complete texts, so it's not at all suited for owners of agencies.


 
José Henrique Lamensdorf
José Henrique Lamensdorf  Identity Verified
Brazil
Local time: 17:38
English to Portuguese
+ ...
In memoriam
Advances in MT seem quite relative Mar 16, 2018

I see many people praising the recent 'advances' after MT switched to 'neural'.

As far as I was able to test and verify, at least when Google Translate deals with EN > PT, this shift to 'neural' represents a HUGE leap BACKwards.

Maybe this 'neural' thing is advantageous for SOME language pairs, at the
... See more
I see many people praising the recent 'advances' after MT switched to 'neural'.

As far as I was able to test and verify, at least when Google Translate deals with EN > PT, this shift to 'neural' represents a HUGE leap BACKwards.

Maybe this 'neural' thing is advantageous for SOME language pairs, at the same time being detrimental to others. I haven't seen any inter-LPs comparison. And there is still the chance that 'neural' works better in one direction than in the other, within the same language pair, as compared to the conventional (viz. previous) system.

Food for research...
Collapse


 


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


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

Neural machine translation offers significant advances with remaining challenges






CafeTran Espresso
You've never met a CAT tool this clever!

Translate faster & easier, using a sophisticated CAT tool built by a translator / developer. Accept jobs from clients who use Trados, MemoQ, Wordfast & major CAT tools. Download and start using CafeTran Espresso -- for free

Buy now! »
Trados Studio 2022 Freelance
The leading translation software used by over 270,000 translators.

Designed with your feedback in mind, Trados Studio 2022 delivers an unrivalled, powerful desktop and cloud solution, empowering you to work in the most efficient and cost-effective way.

More info »