Natural Language text Processing Technology

in all spheres of application for English, German, French and Italian.

Our technology can be applied to any input language using the Latin alphabet, with or without accentuated letters. The output, in case of Machine Translation, is language independent. The output can be in any language in the world. Tested output (target) languages are: Arabic, Chinese, Russian.

Machine Translation is needed often by everybody. Machine Translation from one language into another is never perfect, because we use rigid, structured, mathematically organized programming language to program Natural Language. Natural Language is not a programming language. Natural Language is full of ambiguities on all levels: orthographical, grammatical, morphological, syntactical, semantical, pragmatical. To mention just a few: polysemy, metaphors, metonyms, collocations, co-references, homophones, homographs. Langsoft has found own solutions to these difficulties and has made these solutions public in a number of academic publications and software products.
Machine Translation was always regarded as means to understand, roughly, what was written or said in a language we do not know. Therefore, the translated text needs always post-editing. The myth, that Machine Translation can be perfect, was created by those, who market Machine Translation software programs. Has somebody seen a translation error on a jewel box displaying sample translation? Some companies produce better Machine Translation than other companies. Therefore, only fair competition can bring the best Machine Translation software on the market.

We have produced the following Machine Translation software programs:

for Windows:

- German-English
- English-German
- German-Russian
- French-Russian
- English-Russian
- German-English text-to-voice
- English-German text-to-voice - includes version of SEMCOR
- German-English text-to-video-and-voice
- English-German text-to-video-and-voice
- German-Russian text-to-video-and-voice
- English-Arabic - test version
- English-Persian (Farsi) - test version
- English-Hebrew - test version
- English-Korean - test version
- English-Chinese - test version
- French-Arabic - test version
- German-Arabic - test version
- English-Russian text-to-video-and-voice

SEMCOR was developed to correct automatically errors based on meaning, when the word is spelled correctly, but its contextual meaning is incorrect, for example homophones and homographs. Homophone is a word with identical or very similar sounding pronunciation of two or three words with different spelling and different meanings. The number of homophones in English is around 536, the total number of words involved is around 1740. This is very important for voice-to-voice MT, because Natural Language is full of homophones for example bred - bread, knight - night, cellar - seller, censer - censor - sensor, cheap - cheep, check - cheque, meat - meet, made - maid, plate - plait, death - deaf, shone - shown, etc., for instance, in sentences such as
Charles nights the courageous soldiers.
We dye of hunger.
The rhyme on the apples.
We wine with tears.
The mail swine.
If we add to the existing homophones all mispronounced words ending in -g, -k (sting - stink, log - lock), -t, -d (made - mate, brought - broad), etc., voice-to-voice Machine Translation will be less than 50% correct!
English into plain English MT simplifies the sentence, reduces word ambiguity in context, corrects automatically errors. Thus, it helps language learners and facilitates NLP by setting certain norms on word usage and on expression of meaning needed in Natural Language Understanding.

 
We work, publish books and develop NLP software since 1970