Musings about the Impossible Electronic Dictionary Annie Zaenen Pare, Zaenen@parc.com Sue Atkins and I met over discussions about the polytheoretical, multi- functional lexicon, a bit more than 15 years ago when it became obvious that the computer would have an important impact on how dictionaries were made. The impact was expected to go in two directions. On the one hand, dictionary making would be freed from several of its constraints because more data could be accessed quicker to insure better coverage, size limitations could be overcome and the organization could be more flexible because the information could be accessed in various ways. This much was applicable to human readable electronic dictionaries. On the other hand, it was hypothesized that the information contained in traditional dictionaries would be useful for natural language applications. It was also recognized that the information in traditional dictionaries would not be sufficient and that it would be an enormous task to make adequate lexical databases for NLP starting from that information. The notion of polytheoretical dictionary intended to do away with at least one obstacle that was feared to be in the way of the construction of large reusable lexical databases: the differences in linguistic theories and notational conventions. Of course, we were aware of the fact that linguistic theories in general had rather little to say about lexical matters (see e.g. Zaenen and Engdahl, 1991) but the paradigm for natural language applications at that moment included sentence parsing as an obligatory step, hence the need for lexical resources to be adapted to the various assumptions that parsers might make. Through the years it has become clear that the task was even more monumental than we anticipated but also that, at the moment and most likely for some time to come, the sentence parsing paradigm is of minor interest in natural language applications because, with the availability of the internet for more and more people, the most pressing task has become information retrieval for which parsing is much too slow and unnecessary.