Humanist Discussion Group

Humanist Archives: April 1, 2023, 9:39 a.m. Humanist 36.506 - numbers for words

              Humanist Discussion Group, Vol. 36, No. 506.
        Department of Digital Humanities, University of Cologne
                      Hosted by DH-Cologne
                Submit to:

        Date: 2023-03-31 15:21:40+00:00
        From: James Rovira <>
        Subject: Re: [Humanist] 36.504: numbers for words

Many, many thanks to Michael's and Maroussia's informative responses to my
questions about representing polysemy computationally.

It sounds like, in all instances, these different models can at best
identify different meanings of the same word in different texts, or even
within the same text in different sentences, but I don't see how it's
possible to register different meanings of the same individual word in the
same sentence, or in other words, different meanings of the word in play at
the same time (would this be what Michael meant by Dante's sense?). My
references to Origen, Augustine, Aquinas, and Dante could be a bit
misleading because they're all working with a specific and very limited
hermeneutic tradition, but exploiting different meanings of the same word
in a single use is a common practice in poetry, fiction, and other literary
works, while of course that practice is undesirable in scientific

I've always suspected these limitations were in play and wondered what it
would take to, say, use the OED (Miriam Webster was just close at hand :))
as a set of numbered keys attached to each word identifying a range of
possible meanings (so "father" might have a 1a 1b 1c 1d 2a 2b 2c 2d etc.
attached to it) and then see if the programs could be trained to pick up
multiple meanings of a single use from an expanded context? Ideally of
course we would start with the full OED and then be able to expand our
keys. I think I'm asking for a bit much as this process often involves
interpretive work which is sometimes creative and often requires the
exercise of judgment. We could of course code a few texts this way by hand
and then see if they could train the program to identify patterns on its
own, but then we would be pre-interpreting the text in a senese. It'd get
really involved with poetry, as the program would need to identify end
words and similarly sounding words across different lines as context, not
just words appearing before and after another word.

Jim R

Unsubscribe at:
List posts to:
List info and archives at at:
Listmember interface at:
Subscribe at: