Humanist Discussion Group, Vol. 35, No. 475.
        Department of Digital Humanities, University of Cologne
                      Hosted by DH-Cologne
                       www.dhhumanist.org
                Submit to: humanist@dhhumanist.org


    [1]    From: Tim Smithers <tim.smithers@cantab.net>
           Subject: Re: [Humanist] 35.469: transcribing and translating (72)

    [2]    From: Henry Schaffer <hes@ncsu.edu>
           Subject: Re: [Humanist] 35.467: cautions about digital studies of words (7)


--[1]------------------------------------------------------------------------
        Date: 2022-01-18 08:28:36+00:00
        From: Tim Smithers <tim.smithers@cantab.net>
        Subject: Re: [Humanist] 35.469: transcribing and translating

> On 18 Jan 2022, at 09:05, Humanist <humanist@dhhumanist.org> wrote:
>
>
>              Humanist Discussion Group, Vol. 35, No. 469.
>        Department of Digital Humanities, University of Cologne
>                      Hosted by DH-Cologne
>                       www.dhhumanist.org
>                Submit to: humanist@dhhumanist.org
>
>
<snip>

>    [2]    From: Willard McCarty <willard.mccarty@mccarty.org.uk>
>           Subject: translating? (19)
<snip>

Dear Willard,

You ask "what changed?"

The short answer, I would say, is (so called) Deep Learning
techniques with the massive data sets used to build what are
called the Large Language Models, sometimes talked about here.

The workings of these systems do indeed look impressive, but I
would urge a more careful consideration.  The languages for
which these systems work well are quite limited, compared to
all the languages Humans speak and write.  Is being good at
translating a few languages to a few languages a sign of
dramatic success?  For those who use these few languages,
perhaps yes, for those who don't, maybe not.

Also, we usually only see a few examples of the workings of
these systems at any one time, and are not usually presented
with poor examples of their output.  This, I would claim, is
not a sufficient basis by which to judge them.  Looking at
when and why systems fail to perform well is -- with my
engineering coat on -- a better way to get an idea of how well
a system performs.

Things look Cherry Ripe if you only pick the ripe cherries.

Best regards,

Tim


> --[2]------------------------------------------------------------------------
>        Date: 2022-01-17 07:23:38+00:00
>        From: Willard McCarty <willard.mccarty@mccarty.org.uk>
>        Subject: translating?
>
> The note in yeaterday's Humanist on translating as 'vanity' the
> original Hebrew word sparks a different question. Can anyone give a
> relatively non-technical account of the techniques responsible for the
> radical improvement in translation software over the last decades?
> Consider how well DeepL works, for example.
>
> As some will know -- the story is told e.g. by Yorick Wilks in Grammar,
> Meaning and the Machine (1972, 3-5) -- the Machine Translation project
> failed miserably in the 1960s; out of it came a new or newly invigorated
> Computational Linguistics, and translation dropped out of sight for quite
> a while. What changed?
>
> Yours,
> WM
> --
> Willard McCarty,
> Professor emeritus, King's College London;
> Editor, Interdisciplinary Science Reviews;  Humanist
> www.mccarty.org.uk


--[2]------------------------------------------------------------------------
        Date: 2022-01-18 15:14:18+00:00
        From: Henry Schaffer <hes@ncsu.edu>
        Subject: Re: [Humanist] 35.467: cautions about digital studies of words

Speaking of bigrams, a friend just sent this to me:
https://www.reddit.com/r/dataisbeautiful/comments/s5ozev/gender_of_the_subject_h
e_or_she_of_some_verbs_in/

--henry




_______________________________________________
Unsubscribe at: http://dhhumanist.org/Restricted
List posts to: humanist@dhhumanist.org
List info and archives at at: http://dhhumanist.org
Listmember interface at: http://dhhumanist.org/Restricted/
Subscribe at: http://dhhumanist.org/membership_form.php