A Web-based Advanced and User Friendly System: The Oslo Corpus of Tagged Norwegian Texts

A general purpose text corpus meant for linguists and lexicographers needs to satify quality criteria at at least four different levels. The first two criteria are fairly well established; the corpus should have a wide variety of texts and be tagged according to a fine-grained system. The last two criteria are much less widely appreciated, unfortunately. One has to do with variety of search criteria: the user should be allowed to search for any information contained in the corpus, and with any combination possible. In addition, the search results should be presented in a choice of ways. The fourth criterion has to do with accessability. It is a rather surprising fact that while user interfaces tend to be simple and self explanatory in most areas of life represented electronically, corpus interfaces are still extremely user unfriendly. In this paper, we present a corpus whose interface we have given a lot of thought, and likewise the possible search options, viz. the Oslo Corpus of Tagged Norwegian Texts. A general purpose text corpus should satisfy quality criteria at a number of levels in order to fulfill the needs of the majority of its users. The major users we take to be linguists working in academia, or to some extent in commercial enterprises, such as dictionary publishing. Although corpora exist for many languages, and new corpora are being created all the time, it is surprising how little effort is put into making them fully useful as tools. While creating the Oslo Corpus of Tagged Norwegian Texts, we found that the following criteria ought to be fulfilled: 1) Variety of corpus texts Given that it is impossible to know beforehand what kind of questions linguists (morphologists, syntacticians, semanticists, lexicographers) will want to ask, and to what extent dictionary writers will use the corpus, it is important to ensure that the corpus has a certain size and comprises a variety of genres, and of written standards, if there are any. 2) Variety of grammatical tagging It is vital that grammatical tagging goes beyond simple part-of-speech tagging: In many languages, there is, even within parts of speech, a lot of homonymy that could be disambiguated with a more fine-grained system of tags. Obviously, the tagging also has to be correct. 3) Variety of search options Every feature of the corpus, marked or unmarked, should be searchable. E.g., if the corpus consists of several genres, and the …