150 text samples of approximately 2,000 graphic words each. The samples can be entire articles, excerpts from longer texts, or composite of shorter texts.

The language of the texts is modern written Swedish. The texts must have been first published in 1990 or later. In order to get a balanced corpus with high coverage, the texts are selected and classified according to criteria corresponding to those of the Brown and LOB corpora. The excerption has been done at random, but as far as possible, a coherent and delimited stretch of text has been chosen.

The corpus is almost entirely based on texts that were available in machine readable form, and only to a minor extent relies on scanning or typing in of material.

Clear misprints have been corrected without recording the corrections.

All quotation marks are retained.

All 'soft' hyphenation is removed.

The texts of this corpus fragment are at present segmented into paragraphs. The character @ is used as delimiter.

Each word is tagged with part of speech tagging and morphological information. The principles for the tagging are described in Ejerhed et al.: "The Linguistic Annotation System of the Stockholm-Ume} Corpus Project, version 4.31", Publications from the Department of General Linguistics, University of Umeå no. 32. The tags emanate from an automatic lexicon lookup in the SWETWOL computerized dictionary at the Department of General Linguistics, University of Helsinki, followed by an automatic transduction and a manual disambiguation and correction.

The n attribute of each text in the corpus carries a unique identifying code for the whole text. This code corresponds to an entry in the bibliographical data base. The code is built up from a two-letter code for text type and a three-digit running enumeration within each text type. Where a 2,000-word unit is composite, i.e., consists of several shorter texts, each of those has a lower case letter added to the code.

All words and typographic signs are numbered, and a canonical reference may be made up by concatenating the n value of the text and the word number.