Improving Automatic Translation Software, using Human text-corrections in Automatic Translation

Image via Wikipedia

A few years ago (2003-2004) I made a detailed proposal to several Greek I.T. companies: The proposal was to «improve the performance of automatic translation software» through additional «boot-strapping code» in Prolog, which could learn from the mistakes of Automatic Translation by creating automatic generalizations in NLP (Natural Language Processing) using ILP (Inductive Logic programming).

Fed with human corrections, the proposed software would continually compare human corrections with automatic translation mistakes, so that it would learn (using A.I. techniques such as ILP) how to correct the latter, improving (the final outcome of) Automatic Translation, with time.

The human text-corrections could be stored in a so-called «Translation Memory System» (such as Trados), while the main NLP/ILP program could be a plugin or extension for already-existing Automatic Translation Software (such as SysTran).

Unfortunately, there wasn’t a single Greek I.T. company, willing to risk (my proposed salary of) 10-15 thousand euros (spread over a period of 6 to 12 months) to try out this simple and (in my humble opinion) powerful software-idea. Only one Greek company  (software importers and software developers themselves) told me they might be interested, but that… they would need to see a working software demo of the idea, first! Well, I replied that the proposed initial phase of 6 to 12 months (for which I requested a decent salary) was precisely intended to cover this development phase, to arrive at an «alpha-phase software» implementing the idea. If I could support myself without their help during these few months, I wouldn’t ask for their assistance in the first place! -Instead, I would probably try to sell the «already working demo», or start (once again in life) an I.T. company to distribute the product. But I couldn’t live without any income, during the proposed «initial phase». Well, they immediately lost all interest in the idea, as a result; they wanted to do business with zero risks, zero investment and… zero innovation!
-Ah well, since they were a Greek company, I was not -in the least bit- surprised; always expecting the worst in Greece:

• During the last 3 years, more than 5000 colleagues of mine, Greek I.T. professionals, researchers and scientists have left Greece, emigrating to… saner countries, seeking high-tech jobs; they left because of the extreme narrow-mindedness, imbecility, corruption and complacency that predominate over here:

• Greece is an entire country of -largely- useless, narrow-minded people, most of them living in their own Closed Little World, envying and demeaning each other. Of course, there ARE noteworthy exceptions -typically among those who live (or have lived) elsewhere.
  

• Nevertheless, my proposed «automatic learning software» would be profitable in Greek language automatic translation, since (1) Greek is a relatively rare language, that most A.I./NLP developers do not understand, and (2) there is simply zero competition (for high-tech Greek language software innovations), in a country overwhelmed by so much unbelievable amounts of stupidity, backward-thinking mentality and sheer technological incompetence. Any Greek company willing to invest in this simple idea would become quickly successful, in the Greek language translation market, since my own translation skills in Greek (as well as English) are adequate for the project. The success of the project depends on two types of skill, ideally simultaneous: (1) all the necessary software development skills, and (2) perfect knowledge of at least two languages, for the first language-pair to work properly.

In fact, I wouldn’t even dream of implementing this innovation (in «only 6 to 12 months») for languages I don’t understand, such as German: My idea is guaranteed to produce good results only for Greek-to-English translations (and vice-versa), which is why (to my utmost… disgust) I approached Greek companies (rather than foreign) in the first place. After a while, E-mail sent to companies all over the world was either ignored, or answered with polite replies like «no thanks, we’ re not interested in Greek at the moment».

• Well, if you happen to be an I.T. company or a Greek investor of… rare open-mindedness, just send me an e-mail, to omadeon@hotmail.com, and I’ll be glad to explain it further.

• After the initial alpha-phase, my estimate is another short period (of about six months) to produce a final, fully debugged, nicely packaged shop-ready software product, probably marketed with a nice title like «Automatic Translation Intelligent Companion» (etc).
• However, when your profits-per-month begin to exceed significantly the cost of my salary, I’d be pleased to get a small percentage of your profits, without any salary (a negotiable, modest percentage, somewhere between 15 and 25% of your net income from the proposed product);
  
• offering you unlimited support, as well.
  

• My previous work in NLP is mostly in http://omadeon.com/sourcecode and in http://omadeon.com/sourcecode/hyperlex.html.

Related articles

• Natural language processing fun [via Zemanta]
• Semantic Search With Powerset: Transforming Publications Through Semantic Search Technology [via Zemanta]
• Semantic Travel Search Engine UpTake Launches [via Zemanta]
• Powerset vs. Google: The Completely Premature Head-to-Head [via Zemanta]
• Data, Algorithms, and Powerset [via Zemanta]
• MindMentor, a Robot Psychologist, Does Psychotherapy on the Cheap [via Zemanta]
• Google Translate Becomes the Best Free Online Translator [via Zemanta]
• Inside Innovation at Xerox PARC [via Zemanta]
• Selling and overselling the Semantic Web [via Zemanta]

.