Cart
[Home] [By Thread] [By Date] [Recent Entries]
> > > Provide multilingual mappings for terms and definitions > > > > This is indeed where a tremendous challenge lies. I used to > > be believe > > such mapping was just a matter of effort: lock a few dozen people in a > > room with multi-lingual dictionaries. I now know much > > better. However, > > even in cases where I've dealt with businesses willing to work with > > ontologies (and this is a very sophisticated subset, of course), I've > > found that the old idea is still very entrenched. The > > attitude is "we've > > built multi-lingual apps before. The challenge is no different when > > building a knowledgebase". > > Interesting... Could you make explicit why this is such as challenge? Usually, when you are writing a multi-lingual app, say using GNU gettext, or the many other ML tools available, you are translating self-contained snippets of language. It tends to be little more than a matter of UI testing to determine whether these translated snippets express the message you planned to get across. This common situation is often not too far off from a 1 - 1 mapping. In a knowledgebase, however, on which inferencing or network traversal is to be performed, it is not a 1 - 1 mapping: in many common ontologies, it's not unusual to find vertices with thousands of edges. The fine-tuning that makes a translated expression suitable along one edge usually throws it off for another. In such a case, the mapping itself becomes a mini-graph, which adds a great deal of complexity. Classic network effect. We already know how difficult scalability is to achieve in semantic nets and frame-based systems. The additional dimensions that tend to be introduced by translation edges greatly magnify this difficulty. -- Uche Ogbuji Principal Consultant uche.ogbuji@f... +1 303 583 9900 x 101 Fourthought, Inc. http://Fourthought.com 4735 East Walnut St, Ste. C, Boulder, CO 80301-2537, USA Software-engineering, knowledge-management, XML, CORBA, Linux, Python
|
