Abstract: The experiment reported here aims at identifying the reasons for differing
decisions in identifying and analysing neologisms in a finite press corpus in view
to improving reliability. it follows on from two previous experiments carried out
in differing conditions and with divergent aims. The participants first manually
extracted neologisms from a corpus of french press articles. they then discussed
and corrected their initial findings. After this, they submitted a list of neologisms
extracted and indicated to which matrix each neologism could be attributed. the
proportion of agreement among the participants on the neologisms extracted was
then counted on a scale ranging from complete agreement to none at all.asecond
tally was made of the matrix behind the new word. it appeared that certain participants
favoured certain matrices over others, and a rating of matrices in relation
to the proportion of agreement or disagreement they attracted was also calculated.
this analysis shed light on a number of theoretical points and formed the basis for
a more harmonised approach. the main conclusions have been incorporated into
a set of guidelines used internally and to be reported on in future conferences.
