Re: [Corpora-List] Brown Corpus

From: Jean Veronis (Jean.Veronis@up.univ-mrs.fr)
Date: Fri Jun 17 2005 - 17:03:59 MET DST

Next message: Babis Theodoulidis: "[Corpora-List] RANLP Workshop - Text Mining Research, Practice and Opportunities"

Previous message: Patrick Cassidy: "Re: [Corpora-List] Lexicon with semantic features needed"
In reply to: Adam Kilgarriff: "RE: [Corpora-List] Brown Corpus"
Next in thread: Martin Wynne: "Re: [Corpora-List] Brown Corpus"
Next in thread: Jörg Schuster: "[Corpora-List] Re: Brown Corpus"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi Adam,

Although I agree on the same-size sample design, I am less convinced by
the use of the mean and standard deviation on corpora (as well as
t-score and a few others). The distributions are so strongly skewed that
these measures are probably not advisable. Without getting into anything
too complicated, the median and measures based on it, like the MAD (mean
absolute deviation), and in general what's called "robust statistics",
seem preferable to me.

--j
http://aixtal.blogspot.com

Next message: Babis Theodoulidis: "[Corpora-List] RANLP Workshop - Text Mining Research, Practice and Opportunities"
Previous message: Patrick Cassidy: "Re: [Corpora-List] Lexicon with semantic features needed"
In reply to: Adam Kilgarriff: "RE: [Corpora-List] Brown Corpus"
Next in thread: Martin Wynne: "Re: [Corpora-List] Brown Corpus"
Next in thread: Jörg Schuster: "[Corpora-List] Re: Brown Corpus"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

This archive was generated by hypermail 2b29 : Fri Jun 17 2005 - 17:07:18 MET DST