Sun.13.NOV.2011 -- Launching the German-Language AI Project Although for years it has been our goal to launch a German-thinking AI Mind, we had to advance the state of the art in English-speaking artificial intelligence with MindForth before we could convert the code base of the English AI into a German AI. Because German is so replete with irregular nouns and verbs, we had to implement the proper handling of irregular English words in MindForth before we could visualize doing the same thing in a German artificial intelligence, or "DeKi" (deutsche kuenstliche Intelligenz). We choose the name "DeKi" for our AI code for several reasons, chiefly for being a valid wiki-word for the online documentation, and also for containing the "de" code for "Deutsch" (German) and the "KI" acronym for AI in German.
Yesterday from the Web we printed out twenty-eight pages of a listing of the top one thousand most frequent words in German, so that we may carefully choose a core German vocabulary to be innate in the DeKi AI. Immediately the problem arose, and on a same-day basis we solved the problem, of how our DeKi would recognize an incoming German verb and access the concept of the verb in order to think thoughts with the verb. First we must describe how we deal with the problem in the English AI, before we describe the more complex problem and the more complex solution in German. In English the MindForth AI simply learns a new verb like "play" or "learn" as a complete stem, to which inflected endings like "s" or "ing" may be added by the AI software. In German, especially with irregular verbs, we will need a more sophisticated approach to recognizing each verb as a manipulable concept. Yesterday we hit upon a way to handle the German verb that means "to be" in English. MindForth simply recognizes all present-tense be-verbs as forms of the concept "be". In German we are not able to do exactly the same thing, because the German word for "be" is "sein", which can also mean "his" or "its", and we do not want our auditory recognition AudRecog module to be confused as to whether an incoming German word is a be-verb or an adjective. (Hey, maybe we should call the Russian AI "Eureka", because that word comes to mind right now as we ponder our sudden insight of yesterday.) We suddenly found a solution yesterday with the insight that we could have a be-concept of "SEI" in German, as not the ambiguous, full word "sein" but as the valid stem from which subjunctive forms are created. We can even include "GOTT SEI DANK" ("Thanks be to God") in the German bootstrap "DeBoot" sequence so that the DeKi will have available "SEI" as a lexical item, although a be-verb stem could function deep in the German AI without even being available on the outside surface of the AI. When the German AI receives input of German present-tense be-verbs ("bin; bist; ist; sind; seid"), the auditory recognition software will immediately activate the single concept of "SEI" deep inside the AI. Notice that all five of those forms in German are extremely unique and unambiguous in German, so that the software will not have to jump through hoops of conditional testing before activating the "SEI" concept. Immediately the DeKi German AI may think up a sentence in response by using the grammatical number of the subject and the person of the be-verb to select the proper outgoing be-verb form. To wit, if an American tourist says to the German AI, "ich bin ein berliner", the DeKi can use "SEI" to comprehend and answer, "DU BIST BERLINER". It all works very well as a Gedankenexperiment, Albert, and now we have to do the major scutwork of coding and documenting and uploading the German AI. But we really only want to code enough irregular German verbs and irregular German nouns to get the basic German AI going as a proof-of-concept AI. None of this heute-Deutschland-morgen-die-Welt business. We will put the DeKi up on the Web and let Germany, Austria and Switzerland deal with it. We can retire to the coffee shops of Vienna and Heidelberg and sit in the chairs once occupied by Johannes Brahms or Goetz von Berlichingen. If students want us to solve their AI problems and bring us a nice piece of Bienenstich or Schwarzwaelderkirschtorte, we might take off our Beethoven earphones and give them some help. Mentifex (Arthur) -- http://www.scn.org/~mentifex/AiMind.html http://www.scn.org/~mentifex/mindforth.txt http://www.chatbots.org/ai_zone/viewthread/240/ http://store.kagi.com/cgi-bin/store.cgi?storeID=AMP_Live¤cy=USD ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com