can computers understand human language?
This was the title of a talk I went to today. The lecturer was Prof. Ruslan Mitkov who is one of the prominent figures in computational linguistics. So, what's the answer? Well.. it depends on what do you mean by 'understand' and what do you mean by 'language'. If you are talking about human-like intuitive understanding of unrestricted language with all its complex interrelations with world knowledge, then the answer is no. But if you are talking about the ability to analyse and generate language on various levels and you are letting in the idea of restricting grammar, genre or vocabulary, then the answer is a very enthusiastic yes.
Throughout the talk of course there were all the reminders of how complex human language is. The basic goal of computational linguists is to find solutions to make machines overcome the morphological, syntactic, semantic, pragmatic and discourse ambiguities in language. Automatic anaphora resolution is, of course, one of the interesting challenges which requires more than the knowledge of linguistic rules. So, for example, unless a computer has the sufficient knowledge base about the world (plus the ability to differentiate between humans and, say, ostriches :) ), it would face a problem with assigning the correct reference for "it" in this utterance:
If an incendiary bomb drops next to you, don't lose your head. Put it in a bucket and cover it with sand. (WWII leaflet)
But work on making computers do things with human language has really gone a long way. Among the latest developements is Prof. Mitkov's work on the automatic generation of multiple-choice tests, the automatic identification of cognates, and the automatic simplification of complex sentences for dementia patients. And of course there has been huge progress in the world of machine translation since it started in the forties of the twentieth century.
Anyway, the talk ended with the interesting remark that computers are not trying to replace humans, they just want to help. Well, so far.. ;-)
Mai
Throughout the talk of course there were all the reminders of how complex human language is. The basic goal of computational linguists is to find solutions to make machines overcome the morphological, syntactic, semantic, pragmatic and discourse ambiguities in language. Automatic anaphora resolution is, of course, one of the interesting challenges which requires more than the knowledge of linguistic rules. So, for example, unless a computer has the sufficient knowledge base about the world (plus the ability to differentiate between humans and, say, ostriches :) ), it would face a problem with assigning the correct reference for "it" in this utterance:
If an incendiary bomb drops next to you, don't lose your head. Put it in a bucket and cover it with sand. (WWII leaflet)
But work on making computers do things with human language has really gone a long way. Among the latest developements is Prof. Mitkov's work on the automatic generation of multiple-choice tests, the automatic identification of cognates, and the automatic simplification of complex sentences for dementia patients. And of course there has been huge progress in the world of machine translation since it started in the forties of the twentieth century.
Anyway, the talk ended with the interesting remark that computers are not trying to replace humans, they just want to help. Well, so far.. ;-)
Mai
posted by Mai at 10:26 PM
0 Comments:
Post a Comment
<< Home