|
Custom Search
_here_search_with_expressions_from_this_page ... (_text_below_will_NOT_disappear)
|
v. September 22, 2020
A public knowledge site like this is not the proper location to establish contact with possible investors. This project is described on this public site only because its concept is of general interest for all of us.
Some kind of a universal language learning robot had been conceived several years ago. Here a short description how it worked: The idea processor There is first a ,,virtual machine'', a programmed layer constructed by a minimal instruction set. It is used to place on top of our sequential digital computing a cellular idea processor modeled after the human brain. 'Cells' and 'cell agglomerations' can in some form of well-ordered chaos communicate by exchanging idea-based information pieces of all kind. It is called 'idea processor' and not neural network. The concept of neural networks was not considered as something very useful to simulate on computers like ours and like in our time. Language data is ,,learnt'' into the cellular structure of this idea processor. There is also a universal grammar which tries to organize the data conforming to habits known from the human brain. This concept is evidently very different from usual concepts of machine language translation. Current concepts of machine language translation rely heavily on the analysis and constructing concept. This is the way like languages are still learnt in most schools. The problem with this is, human brains do not work like this. This explains the well known limits of machine language translation. It also explains why grammar-based language learning in schools never in human history worked properly. Quality of machine translation did not enough advance from the first concepts half a century ago and the level obtained now. The reason for this has been explained here. There is no chance to reach the goal by step by step improvements. On a road which does not lead to Rom, you will possibly with many steps come closer to Rom, but you will never reach Rom.
This concept is modeled after the human brain. This is which enables the difference in comparison with traditional language translation. OMNI-TRANS somehow ''thinks'' the content while it translates. OMNI-TRANS (alias Hyper-Blabla) could be run on Internet. servers. This would be the most efficient and most probable form of availability. A beta version of OMNI-TRANS (popular: ''Hyper-Blabla'') had been distributed 1996 in France. The irrational aspects of French technology protectionism finally disturbed the project. Beginning distribution of OMNI-TRANS was cancelled instantly by its developer. All software packages were called back from resellers. (Only approximately 10 copies of French/German had already been sold for the equivalent of approx. 30 Euro / 30 USD per item.) Read a bit more about the story here: mot7.org
It was bound to the very old Microsoft PC-DOS operating system. The development of the ''virtual machine'' would have to be done again, now possible within the current multiple computing power. Some parts have already been implemented on top of LINUX (portable - would be able to run on all computers like Microsoft Windows, MAC, with some compatibility tools also on mobile phones). Typically, the language translation software should execute on Internet servers of a translation service site, hence only in online mode. Evidently, in this case all Internet-capable devices would be able to benefit from the translations.
Simultaneous translation during phone calls would be possible when using VOIP. But simultaneous translation was never easy, even not for human translators. OMNI-TRANS can not be as good as human translators for everyday conversation. This is because a robot is not taking part in social life. A robot has no human-like identity and reality. A robot might be better then humans for scientific text and might be sufficient competitive for handbooks. But a robot will not be competitive with bi-lingual humans for human conversation where participation in human life and trends and emotions is vital. So translation of poetry, too, will remain in the hands and minds of humans. These restrictions may be valid for the next hundred years or so, What might be afterwards? ..... do you really care? What is your age?...
Otherwise it was at least of interest to describe how computerized language translation could be if it would be more than ''machine'' language translation.
|
The reactivation of this past project is possible. This is not a project for rapid success. The first version might be available within 12 months. Several years will be needed for relative completion. THE PROBLEM: The quality of foreing language translation did not much improve from 1960 until 2005. The software systems were based an grammar, not on language constructs. - Since 2005 rapid improvement has been obtained. On the other hand: Basic limitations of these improved concepts can already be analyzed. Current prevailing concepts for machine translation are based on correlation statistics (with some refinement). This enables to obtain automatic language translation for many language pairs with a reasonable effort. This is the prevailing method since approximately 2005. This method of the statistical approach has been studied here. The opinion is that this method can never reach a quality level close to perfection. The quality level is high (but not yet tolerable) between English and French. The research workers apparently mostly knew these two languages. Apparently some refinement was added for the French language. This refinement was apparently based on lingustics, not on statistics. The quality level is also possibly high related to languages with a basic structure similar to the English language. This is perhaps the case for the Chinese language. The quality level is low for languages for which the statistical method is not sufficient. The translation quality is still too low for the German language. The translation errors are mostly the evident result of the statistical method. Improvement would require to add linguistics. Read here below why linguistics will also not obtain perfection. Probably the quality level is low for all languages which are a natural melting pot and synthesis of various very similar regional languages. Therefore the quality level of the statistical concept will probably never be good enough for the Arabic language, the Russian language and the German language. Such languages followed a natural evolution path. They are therefore full of irregularities. These languages are full of complex structures to express subtle aspects not by vocabulary but by structures. These languages are strongly related to the archetypes of the specific culture. They are full of ambiguities. This problem type can probably not be mastered by correlation statistics and not by linguistics. The past prevailing concepts for machine translation using linguistics. The use of correlation statistics was and is a significant progress of efficient results. The past prevailing method (1960...2005) was based on grammar and vocabulary. The managers and scientists thought that the research workers in the field of language would be the best choice for the development of software for language translation. This is the same error like for past prevailing language teaching methods in schools. Scientists are good for science. This does not mean that the science of linguistics is helpful for the reality of teaching languages or for the reality of developing software. The translation software based on linguistics was a significant progress, compared to the option to have nothing. The results are important and will be useful for all future software development in the field of linguistics. The disadvantage is that the human language brain does not obey to the limited rule set of linguistics. Linguistics are useful to mirror structural features of language reality. They are nor sufficient to construct language reality. The major disadvantage of language translation software based on linguistics is the lack of efficency. A significant human labor and financial investment was and is required for each language pair. Language translation software of this type covered during half a century always only a small number of some most spoken or otherwise important languages. The quality progress during half a century was not extremely convincing. There is no significant quality progress to obtain without a significant investment volume of human labor, hence of money. This is an approximation problem. On the road to perfect language translation, the next steps are always smaller and become more and more expensive. You will always run behind the carrot and you will never have a chance to grasp it. Photo: how to master language barriers. (© Ohto Kokko, Finland, GNU Free Doc.Lic.v1.2++) THE SOLUTION: OMNI-TRANS tries to overcome the described shortcomings. OMNI-TRANS included already in its first distributed version (1997) a totally different concept. The concept tries to be based on the mathematics of language. The concept tries to mirror the mathematics governing the human language brain. This is based on the opinion that some kind of complex mathematics is a built-in feature of the language brain. The origin of this concept is the observation how native speakers learned their own language. Children learn certainly with methods which are not based on linguistics and not based on correlation statistics. The concepts of linguistics and of correlation statistics are only involved in some auxiliary role of a slave. The governing master concept is different. This governing master concept is what OMNI-TRANS tries to translate into software. Here is no knowledge of any other product creation project for this kind of language translation. The approach is in fact of a high complexity degree. OMNI-TRANS has been developed for auto-learning languages like a chield. It is working close to the human brain. The concept is based on idea processing in an idea network. Thinking is the master, like in the human brain. Language is the slave, like in the human brain. Linguistics and correlation statistics are involved but are not dominating the process. A first version was in public distribution in France in 1997. It was withdrawn one month later from the market due to some external reasons. (In France, public adminstration is extremely hostile to innovative small businesses. The project will never again be exposed to the risks of this hostile environment.) Since then, the project was suspended, due to the evident progress of other projects. During these years, the opinion was that these other concepts would be similar to that of OMNI-TRANS. A more detailed analysis of other concepts was done for the first time in early 2011. The result is listed here above. It means that there is still the need for something like OMNI-TRANS. Something like the OMNI-TRANS concept is perhaps the only way to obtain one day a software for really perfect translation of spoken language. The road will be long. In the long run, the project has to include content analysis in the background. This means some kind of "thinking". It would also require image-based environment analysis and sound-based analysis. See for these aspects other projects like the: "SPHINX Classificaction system" and "Encyclopedic approach to information access with included quality selection.". In addition, it would require an idea simulator working similar to the human brain. This last project with ongoing internal software development is at present not offered for investors. It is just research work and science. There is no way to earn money from it. It is now necessary to compete with the increased quality competition obtained by software translated with correlation statistics. A significant initial investment is therefore now required. This project is at present in preliminary examination by a possible investor. The reservation period is two months. As usual, most preliminary examinations do not result in an agreement. FUNDING / REQUIRED: Up to public availability: 2 million USD. - Participation: 10 (or more) % from this. FINANCIAL OFFERS: EBB-JJA-TRANS Your msg.(with code above) to forwarder: ok @ fin7.com (This is a module of a more comprehensive overall project. But it can also be financed independently and used.)
_________________ Photo: The monster truck (J.Powers 2004, Creat.Comm.Lic.2.0) THE PROBLEM: EXAMPLE: On prof7.com / start page, search for the title line: "AHA7 FIN MONSTER". This is an attempt to create with software translation a multilingual encyclopedia of the vocabulary of the financial markets. A basic linguistic problem is involved. The vocabulary for financial subjects has many ambiguities. This is frequent in all cultures and has complex reasons. - The question was, is there a way to detect the translation errors and to rectify them, all this by software? INFORMATION: INFORMATION: _EN_ =in English language: prof7.com / start page - headline: "AHA7 FIN MONSTER" _DE_ =in German language: aha7.com _FR_ =in French language: aaazzz.com THE SOLUTION: A suitable algorithm has been developed. It can detect faulty translations and can rectify them. (It has not yet been applied to the translations which are described here above.) FUNDING / REQUIRED: Required is patenting, then licensing. - Examples of the lucrative patentability of software algorithms: MP3, encryption, etc.. FINANCIAL OFFERS: EBB-JJA-TRANS Your msg.(with code above) to forwarder: ok @ fin7.com (This is a module of a more comprehensive overall project. But it can also be financed independently and used.) |
|
-ana-pubcc-pha11609# D=20922 F=ebb-jja-trans-en.htm |