The two free services I know of are Google Translate and Yahoo! Babel Fish (there are probably others so if you know of one that does better than these, please leave a comment). I took a paragraph from an article on the French language on Wikipedia and copied it into each service. Below is the orginal paragraph, followed by the Babel Fish translation and then the Google (Depending on the character set used, the original paragraph may have strange characters displayed. I apologize if that occurs).

I don't speak French so I don't know how close the translation is to the original, but it does appear things are moving along and perhaps MT is at a point where is useful, if not perfect.

Original Wikipedia Paragraph

Le français est une langue romane parlée principalement en France, dont elle est originaire (la « langue d'oïl »), ainsi qu'au Canada (principalement au Québec, dans le nord du Nouveau-Brunswick et dans l'est et le nord-est de l'Ontario, et aussi au Manitoba, en Saskatchewan, en Alberta, et en Colombie-Britannique), en Belgique (en Région wallonne et en Région de Bruxelles-Capitale) et en Suisse (en Romandie). Le français est parlé comme deuxième ou troisième langue dans d'autres régions du monde, au total dans 51 pays du monde ayant pour la plupart fait partie des anciens empires coloniaux français et belge.

Yahoo! Babel Fish Translation

French is a Romance language spoken mainly in France, in which it is originating (the “language d' oil”), thus qu' in Canada (mainly in Quebec, in the north of New Brunswick and l' is and the North-East of l' Ontario, and also in Manitoba, in Saskatchewan, in Alberta, and a Colombia-British), Belgium (in Walloon region and Area of Brussels-Capital) and in Switzerland (in Romandie). French is spoken like second or third language in d' other areas of the world, on the whole in 51 countries of the world for the majority having belonged to the old French and Belgian colonial empires.

Google Translation

The french is a Romance language spoken mainly in France, where it originates (the "langue d'oïl") and Canada (mainly in Quebec, northern NB and in the east and north - eastern Ontario and Manitoba, Saskatchewan, Alberta and BC), Belgium (Walloon Region and Brussels-Capital Region) and Switzerland (Romandie). The french is spoken as a second or third language in other regions of the world total in 51 countries around the world have for the most part former colonial empires french and Belgium.

Aloha!

Don't get me wrong! Machine Translation does serve a purpose, but it does require supervision as well!

Anyway, just so you know, this is a restaurant ... where the main dish is probably "translation errors with rice".

• I happen to be a computational linguist, among other things, and I disagree with Cerf's assessment of the present and near-future success of automatic translation -- you can't translate correctly if you can't think. Computers can't think, and we won't know how to make them think for quite some time.
• His assessment of the impact of mobile computing is spot-on, though, and it's a point that's often not on the radar in western industrialized countries with hot and cold running PC's.

You can use this code on your pages or articles that you think they are Machine Translation Ready:

Here you have other variations if you prefer, just rename the image in the code accordingly:

Source: ELBES Multilingual Communication - Machine Translation (Multilingualism.org) ]]> http://lostincode.wordpress.com/?p=106 Mon, 11 Aug 2008 17:14:53 +0000 ChrisCicc http://lostincode.wordpress.com/?p=106 The Microsoft Research Machine Translation (MSR-MT) Team today annouced they have released the new Office Translator plugin for Office 2003 and 2007.

I have yet to test out the tranlation capabilities, but when I do I'll post here!

In the meantime go read the instructions on how to install it manually now, or wait for the Windows Update!

Read

Blogoscoped reports that Google is preparing to launch Google Translation Center, a new translation tool for freelance and professional translators. This is an interesting move, and it has broad implications for the translation industry, which up until now has been fragmented and somewhat behind the times, from a technology standpoint

Google has been investing significant resources in a multi-year effort to develop its statistical machine translation technology. Statistical MT works by comparing large numbers of parallel texts that have been translated between languages and from these learns which words and phrases usually map to others — similar to the way humans acquire language. The problem with statistical MT is that it requires a large number of directly translated sentences. These are hard to find, and because of this SMT systems use sources like the proceedings from the European Parliament, United Nations, etc. Which are fine if you're writing in bureaucrat-speak, but aren't so great for other texts. Google Translation Center is a straightforward and very clever way to gather a large corpus of parallel texts to train its machine translation systems.

Part machine translator and part translation memory (a sort of search engine for translation that helps translators to recall translations), GTC will help translators by providing a free, global translation memory, and in turn drive costs down by reducing the amount of work needed to complete a text. It will help Google by providing an excellent source of high quality parallel texts that can be fed back into the statistical translation systems.

If Google releases an API for the translation management system, it could establish a de facto standard for integrated machine translation and translation memory, creating a language platform around which projects like Der Mundo can build specialized applications and collect more training data.

On the other hand, GTC could be bad news for translation service bureaus — especially those that use proprietary translation management systems as a way to hold customers and translators hostage. Most translation bureaus aren't really technology companies and aren't very competent at building quality software. Google Translation Center fills a void in the translation tools market that was created when the few independent companies, such as Trados, were acquired.

For freelancers, GTC could be very good news; they could work directly with clients and have access to high quality productivity tools. Overall this is a welcome move that will force service providers to focus on quality, while Google, which is competent at software, can focus on building tools. Google has a pretty mixed track record with consumer-facing services outside its core search business. But if it positions itself as a neutral service provider, it could enable projects like Der Mundo and others to create powerful and easy-to-use translation services for a broad range of industries.

Translation management is more complex than it appears, with different practices in different industries. If you're translating a news story, you want minimal cost and fast turnaround time (publish early, correct often). If you're translating a product spec sheet, you're willing to spend more to have it done right before it goes to press. Google would be smart to position GTC as a utility for translators and to encourage service bureaus to standardize around it, much as it did around earlier tools like Trados, and much as it has done with their keyword ad business. That strategy would also eliminate a potential conflict of interest, as translation professionals are understandably wary of contributing to something that could put them out of work, as well as avoid channel conflicts with partners who will be their best advocates in selling to various clients.

While it's my guess that Google has no intention of directly monetizing the service (charging a commission on transactions it brokers would expose Google to a billing and payment disbursal nightmare), the R&D value of collecting millions of parallel sentences in every language pair imaginable is indisputable, and it will pay off in unforeseen ways. So, my guess is Google will make this a free tool for the translation industry to use, and it will figure the money part out later. It can afford to be patient.

Translation is a very difficult problem. If it weren't, it would have been solved a long time ago. I remain convinced that a multilingual web will be a reality in a short time, and that a menagerie of tools and services will emerge over the next few years — some geared toward helping translators, some toward building translation communities, and others that make publishing multilingual sites and blogs easy and intuitive.

As these emerge, the web will begin translating itself, and within a short time, we'll be able to read content from sources worldwide just as we currently explore the web in our own language today.

"Tradd.us is an innovative and creative approach to text translation. Powered by Google translator and a range of other services we are able to translate, analyze, identify and contextualize important information from your content, in a very quick and organized way."

To check it out, first you need to sign up here. Soon after your will receive an invitation by e-mail.

The tool uses semantic analysis to select interesting pieces of the translation and divide them into tabs, under which you get extra information, such as synonyms. It also helps you to find relevant information automatically by showing the best ranked sites for those therms selected, although it also shows paid links, apparently.

It seems to be an interesting tool, unfortunately, it didn't prove to be much useful for me at the moment, as the only available choice is translations from English into another language, and I would be more interested, work-wise, on the other way around. But let's experiment, translating the quote above into Portuguese:

Tradd.us é uma abordagem inovadora e criativa ao texto da tradução. Powered by Google tradutora e uma série de outros serviços que são capazes de traduzir, analisar, identificar e contextualizar as informações importantes a partir do seu conteúdo, de uma maneira muito rápida e organizada

Not too bad at all - only one mistake in the first sentence, an untranslated "Powered by" in the second followed by a couple of little syntax mistakes. If I were translating it, it would be:

Tradd.us é uma abordagem inovadora e criativa para tradução de textos. Desenvolvida pelo Google Translator e uma série de outros serviços, somos capazes de traduzir, analisar, identificar e contextualizar informações importantes encontradas no seu conteúdo, de uma maneira bem rápida e organizada

[... continue ...]

this idea crossed my mind when i was thinking back to how advanced tts has become in the past two years (especially those developed by microsoft's competitors) and when i was reminiscing back to the fact that it IS entirely possible to teach a computer how to speak a (human) language fluently, just impossible to give it its own tongue and a mind to communicate with other humans (and by communicating i mean the exchange of semantically and logically irrelevant language, like the ones only humans are capable of engaging in).

also the fact that a computer is supposeldy incapable of independent bias led me to believe that in the future translation and interpretation will all be outsourced to computers and machinery, unless the field of computational linguistics hits a huge brick wall and fails to progress from now until the end of time.

so in hypothesizing such an occurence of the future, i myself deduced the possible inner mechanism of a computer/program/machine capable of such a feat, and it scared me to think that such a machine could easily be built should the idea catch the attention of interested parties, or to think that there may already BE developers/inventors who could easily build such a mechanism... which led me to come up with a little blueprint of the machine of my own...

these are the key components of my "real time translator":

1. speech to text
using speech-to-text technology, the machine will acoustically record and analyze the speech being spoken and transfer it into data, most likely in some form of text. http://www.brothersoft.com/downloads/speech-to-text.html is an example of speech to text technology being developed all around the world.

2. sentence breaker/pos tagger/word breaker
after the speech is transformed into analyzable data, the spoken speech is analyzed by a sentence breaker which given its knowledge/background in the syntactic structure of the language being spoken, breaks down the speech cluster into sentences. after the speech is broken down into simple sentences, each word is separated and given a "part of speech" tag, depending on the word's placement within the sentence and the context of the sentence. The NLP project has demos for POS taggers and it is a widely known fact that Nuance and Microsoft have both been working on sentence breakers/word breakers for a long time now.

3. lexicalization
after each word has been broken down and tagged with a part-of-speech, the word is then referenced to the main language lexicon, which is basically a huge dictionary that stores information regarding how each word is pronunced, its frequency in usage within the language, how the word is used in different parts of speech if such information is applicable and so on. after such information is acquired from the main language lexicon, it needs to be then cross-referenced to a lexicon containing the same information in the target language so that "translation" can take place.

4. pos tagger/syntax builder
now that the "translated" data is available in the target language, another pos tagger needs to be applied in order to correctly label the new data, which will then be fed through a syntax builder in order for it to be correctly and accurately formed into a logical sentence in the target language.

5. text to speech
once the sentence is completely translated into the target language and is found to be syntactically and semantically accurate, the sentence then needs to be fed through a text to speech engine which will then relay the speech back to the targeted audience. text to speech can be found everywhere in the modern computer age, anywhere from global navigation systems, registry id calls, and even in windows pc's which comes standard with a mediocre version of it in every copy (if you're bored, go to accesories > accessibility options > text-to-speech)

the understood difficulties of this project are numerous and tantamount in scale: the lexicon will have to be updated on a regular basis to account for new words, terms, and definitions; machine translation would mean that translations will often lack variety and be monotonous in nature; the problem of how to set the machine to deal with terms and data that may not be within the lexicon (i.e. names of people, location, new things that may seem obscure); the irregularity of language that will most definitely throw the machine off course; and also the huge amount of processing power required would make instant translation/interpretation very hard or almost impossible.

but as mentioned before, the benefits of such a machine would be endless as it would bridge countless gaps and holes that are duly formed because of language barriers, although it could effecitively mean that what was once a proud oral tradition of human kind will now be lost and permanantly outsourced to hearltess machines.

oh, and i'd be out of a job too, but that's beside the point...

First official release (ver. 0.1) of Anubadok was an experimental release which mainly served as a proof-of-concept for an open-source English to Bengali machine translation system.

With the release of version 0.2.0, I am glad to upgrade its official tag from "an experimental software" to "a software under development" with clear-and-specific implementation targets. However given the nature of the project, there are no specific time-frames for future releases. Further, given machine translation is considered an open research topic in Computational Linguistic, you should expect to see some surprises ;) even for well implemented situations. Specially, if you are comparing results of machine translations with human translations.

In English, there are four types of sentences: Declarative, Imperative, Interrogative and Exclamatory. These sentence types further fall into four basic sentence type: Simple, Compound, Complex and Compound-Complex.

The table below gives approximate status of implementation for each sentence type in the current release and inversely it gives the targets for future implementations.

W: Well implemented
M: Moderately implemented
N: Not/Not-well implemented

Anubadok does not yet have any code to handle Complex or Compound-Complex sentences, not even moderately. This is where next push for development is needed.

Few other salient features of this release:

• The execution method of Anubadok system has been re-written. Anubadok itself has been implemented as Perl module. This means one can now access Anubadok in a Perl program directly by including Anubadok libraries (Perl modules) or in any other program by using appropriate Perl module wrapper.
• The notion of "testsuites" has been introduced for Anubadok. For a given English sentence, it compares a machine translated sentence with the expected Bengali sentence. This is quite an useful tool while adding new features or doing some experimentations as it would ensure that already implemented algorithm are not affected.
• Anubadok system can now handle several kinds of input documents including plain text files, any XML documents, HTML files with in-line javascript, CSS. Further, as earlier, it is capable of translating Portable Object (PO) files directly.
• Anubadok packaging has been completely reorganized to ensure that it has the basic structure of a standard Perl package. Consequently, Anubadok can be installed following the method of standard Perl module installation.
• Anubadok-0.2.0 comes with an updated dictionary having 15K+ entries in its database. This is almost double the number of entries it had in 0.1 release. Credit for this goes to all the contributors of Ankur English to Bengali dictionary project. Anubadok's dictionary are now updated regularly using database dumps of Ankur E2B dictionary.
• Anubadok has now moved to its new website hosted by SourceForge.

  http://anubadok.sourceforge.net

  Latest source codes of Anubadok can be downloaded from the "trunk" branch of its SVN repository.

• Anubadok Online, the online interface to Anubadok system, has been upgraded substantially. It runs directly using SVN version of Anubadok engine. User contributed new entries though this interface are submitted automatically to Ankur E2B dictionary project.
• A brief document is now available for download as a PDF file from its website. It describes the internal working and the algorithm used by Anubadok system by considering specific example sentence.

1. The size of input text.

2. The text must be well formed.

3. If the lexicon has to built, that could be treated as atomic or complex.

4. The treatment to the text can be divided into semantics, lexicon atoms, grammar and context.

5. The loss of information during translation.

6. A flat layer between two natural language... a language

MIP (Managing Intellectual Property) is a trade journal out of London that does a good job of providing global coverage. They have a free newsletter called MIP Week, which is perfect for people like me, who want to stay abreast but are too cheap and lazy to read the whole magazine.

Yesterday they had more on a story that they originally reported In April -- specifically, a translation breakthrough for Community Patents. The idea is to start granting European patents, without demanding that applicants provide translations into all 23 official languages. People wanting to read the patent could then use a proposed MT (machine translation) system. Interestingly, MIP reports that the translations would have no legal value, and that for the 1% of patents that are litigated (man, that's a lot of litigated patents) human translations would prevail.

I can certainly imagine some juicy courtroom arguments when they start litigating patents without predetermined translations. It also makes me wonder about enforcement. If I spent money developing something that my research (reading MTed patents at the EPO) told me is not protected, and then I got sued for infringement because it turns out the MT translation was inaccurate, I would certainly feel mistreated. If my due diligence is expected to extend to procuring an accurate translation by myself, then the EPO is transferring translation costs to industry. If, on the other hand, due diligence doesn't extend that far, why should I be penalized?

In any case, it should be fun. It makes me wish I were a European legal translator.

Given that this blog is about named entity recognition (NER), itself an NLP application, we would be biased at including NER to this list. As such, we've excluded ourselves from the chart-toppers ;)

#5: Chat bots

"HELLO, MY NAME IS DOCTOR SBAITSO.

I AM HERE TO HELP YOU.
SAY WHATEVER IS ON YOUR MIND FREELY,
OUR CONVERSATION WILL BE KEPT IN THE STRICTEST CONFIDENCE.
MEMORY CONTENTS WILL BE WIPED CLEAN AFTER YOU LEAVE,

SO, TELL ME ABOUT YOUR PROBLEMS."

The first time I chatted with Dr. Sbaitso, I was about 12 years old. Probably more than anything else, it has influenced my career path. Since then, chat bots such as ELIZA, A.L.I.C.E. and Jabberwacky propelled the art of conversational robots, leading to Automated Service Agent applications (see NextIT)

For its lasting impact on generations of NLP developers, and for the interesting improvements that ensued, Chat bots rank #5.

#4: NLP-based search engines

Ask Jeeves pioneered it, Powerset redefined it, but we are all somewhat skeptical when it comes to beating Google's classic vector space models and ranking techniques., Do we really need shallow NLP parsing to answer "When did Einstein die," or will statistical fact extraction suffice?

Though it is the Holy Grail of NLPers, it has not yet surpassed current information retrieval techniques. As such, NLP-based search engines rank #4.

#3: Speech recognition

Microsoft and Ford just teamed up to develop in-car speech recognition. But they forgot to include Electronic Voice Alert, a feature of mid-80s luxury Chrysler cars!

In all seriousness, automatic speech recognition (ASR) is a vital application for hand-free computing (for disabled persons or for certain circumstances, such as driving), and transcription. It is also poised to revolutionize audio-video content retrieval.

For where it came from, and for where it's going, ASR ranks #3.

#2: Machine translation

"It is apparent to me that the possibilities of the aeroplane, which two or three years ago were thought to hold the solution to the [flying machine] problem, have been exhausted, and that we must turn elsewhere." - Thomas Edison, inventor, 1895

The "heavier-than-air" problem that once plagued flight technology is probably the best comparison we can make to AI and machine translation (MT). It was long believed that MT would require a completely automatic understanding of human language before a resolution finally came. But today's Google and Government of Canada systems surpass human translation abilities (can you translate from French to Chinese? Not me.) Their good level of precision makes them useful in many applications.

People are constantly pinpointing these systems' shortcomings, but nobody would contest their second-place ranking on this list.

#1: Knowledge discovery in texts

Have you ever heard of software that finds new relationships and interactions between genes, proteins or cells? By mining large collections of scientific literature, NLP agents can discover and highlight novel and surprising knowledge.

What makes knowledge discovery so promising is the hope that, in the near future, we may monitor all these documents that are just too abundant to be processed manually. Early forms of knowledge discovery, such as data mining, are already used for Business Intelligence (BI) and outside the NLP world, examples of machine-made inventions already exist.

As a form of technological singularity, and as an emerging field of research for NLP, knowledge discovery gets first place on this list of top NLP applications.

It matters how well judges can score MT output, because that is the evaluation standard by which automatic metrics for MT evaluation are judged.  The better an MT metric correlates with how human judges would rate sentences, the better.  This not only helps properly gauge the quality of one MT system over another, it drives improvements in MT systems.  If judges don't correlate well with each other, how can we expect automatic methods to correlate well with them?  The standard practice now is to normalize the judges' scores in order to help remove some of the bias in the way each judge uses the rating scale.

Vilar et al. (2007) propose a new way of handling human assessments of MT quality:  binary system comparisons.  Instead of giving a rating on a scale of 1-5, they propose that judges compare the output from two MT systems and simply state which is better.  The definition of what constitutes "better" is left vague, but judges are instructed not to specifically look for adequacy or fluency.  By mixing up the sentences so that one judge is not judging the output of the same system (which could introduce additional bias), this method should simplify the task of evaluating MT quality while leading to better intercoder agreement.

The results were favorable and the advantages of this method seem to outweigh the fact that it requires more comparisons than the previous method required ratings.  The total number of ratings for the previous method was two per sentence:  O(n), where n is the number of systems (the number of sentences is constant).  Binary system comparisons requires more ratings because the systems have to be ordered:  O(log n!).  In most MT comparison campaigns the difference is negligible, but it becomes increasingly pronounced as n increases.

What would be interesting to me is a movie recommendation system that asks you a similar question:  which do you like better?  Of course, this means more work for you.  The standard approaches for collaborative filtering would have to change.  For example, doing singular value decomposition on a matrix of ratings would no longer be possible when all you have are comparisons between movies.  Also, people will still disagree with themselves (in theory).  You might say National Treasure was better than Star Trek VI, which was better than Indiana Jones and the Last Crusade, which was better than National Treasure.  You'd have to find some way to deal with cycles like this (ignoring it is one way).

References

Vilar, D., G. Leusch, H. Ney, and R. E. Banchs. 2007. Human Evaluation of Machine Translation Through Binary System Comparisons. In Proceedings of the Second Workshop on Statistical Machine Translation. 96-103. [pdf]

1. Eliminate need to recharge batteries on wireless devices
2. Improved parallel processing (at the PL and OS levels)
3. Gesture detection
4. Speech-to-speech machine translation
5. Long term persistent storage
6. 100-fold increase in programmer productivity
7. Identifying the financial consequences of IT investment

I think numbers 1-3, 5, and 6 are almost certainly doable (though they all lie outside of my expertise).  Number four will at least make very long strides towards being widespread and easy to use.  I seriously doubt it will be perfect (and by perfect I mean as good as a trained  translator).  Number 7 I have no idea about, but getting management to understand the exact benefits of IT has been elusive for the past twenty-five years.  I doubt IT managers even have that kind of understanding about it.  There are so many variables.  As humans are made to become more and more slaves to their corporate overlords (i mean protectors), perhaps prodcutivity will become more predictable.

In a funny case of mistranslation, it seems like someone had been handing out bilingual "IT'S GONNA HAPPEN" hand held signs to Cubs fans as part of opening day promotions (this is the 100th season since their last World Series win for the Lovable Losers, and they have a pretty good team). But the actual Japanese printed on the signs "偶然だぞ" actually means "You Were Lucky".

So, it's almost ironic that Fukudome had a great night while the home fans were unintentionally telling their newest import that he was just lucky. Hahaha, machine translations. That's why us translators will still have work for the foreseeable future.

Google Tranlsate gives "偶然だぞ" for "It's gonna happen" and I guess that's what the person used without checking it first with any number of Japanese speakers in Chicago. It's on par with various Engrish found around Asia and misused Chinese characters in the West. It's at least kinda funny though.

Created at Princeton University for research in Machine Translation. An offline version can be downloaded from the official Princeton University website.

If your interest is piqued too, the listings are for:

1. Research Scientist in computational linguistics
2. Program manager
3. Software Engineer

And, of course, salary ranges are not provided.

The subtree metric is a fairly straightforward idea. You begin by parsing both the hypothesis and the reference sentences using a parser like Charniak or Collins to get a Penn TreeBank style phrase structure tree. You then compare all subtrees in the hypothesis to the reference trees, thresholding the number of matches by the best match in the reference trees. The formula is given below:

The tree kernel metric uses convolution kernels discussed by Collins and Duffy (2001). For the specifics of this method, I refer you to the respective papers (and I may post on it at a later date), but the general idea is that you can transform structured data (a tree) into a feature vector by using the kernel trick. Finding all subtrees of a tree can be exponential in the size of the sentence, which would make computation infeasible for large sentences. The kernel trick lets us operate in this exponentially-high-dimensional space with a polynomial time algorithm. Once we have constructed the feature vectors for the hypothesis and refernece trees, we can score them with their cosine similarity:

H(T1) and H(T2) are vectors with non-zero values for subtrees (dimensions) that appear in each tree, so the dot product of the two is the number of subtrees in common. The score is computed as the maximum cosine similarity between the hypothesis and the references.

Finally, the headword chain metric (HWCM) relies on dependency parses, which I touched on in my previous post.

The HWCM score is calculated just like the STM except by comparing headword chains. The difference between the HWCM and the dependency version of the STM is that STM considers all subtrees whereas HWCM only looks at direct mother-daughter relations (no cousins or sisters).

References

Michael Collins and Nigel Duffy. 2001. Convolution kernels for natural language. In Advances in Neural Information Processing Systems.

Ding Liu and Daniel Gildea. 2005. Syntactic Features for Evaluation of Machine Translation. In Proceedings of the Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization at the Association for Computational Linguistics Conference 2005, Ann Arbor, Michigan.

Karolina Owczarzak, Josef van Genabith, and Andy Way. 2007. Labelled Dependencies in Machine Translation Evaluation. In Proceedings of the Second Workshop on Statistical Machine Translation, pages 104-111, Prague, June 2007.