Language makes us human. But there is an interesting asymmetry in our willingness to ascribe linguistic capacities to non-humans: animals are seen as having none, whereas computers may well master it according to many. What curious conception of language makes this asymmetry possible? And what do Descartes and Turing have to do with it? Notes from a new essay about language between animals and computers.

Mark Dingemanse

Language is what makes us human: one of those things, perhaps *the* one
thing, that sets us apart. But there is an interesting asymmetry in our
willingness to ascribe linguistic capacities to non-humans: animals tend
to be seen as having none, whereas computers are increasingly thought to
have mastered language. This asymmetry is the focus of [a recent
essay](https://doi.org/10.1075/avt.00095.ras) I co-authored with a range
of people, led by Marlou Rasenberg.
^[1](#footnote_0_6861 "Our piece won shared second prize in an essay competition on Big Questions in Linguistics organised by LOT, the Netherlands Graduate School for Linguistics."){#identifier_0_6861
.footnote-link .footnote-identifier-link}^

People will of course grant that other animals have communication
systems, too; they're just not language. Some people feel they have
meaningful interactions with their pets; that too, consensus says, is
not language. In short, scientists are generally careful, frugal even,
when it comes to ascribing linguistic capacities to non-human animals.

With computers on the other hand we are not so cautious. Already in the
1960s there was the idea that a computer should be able to learn
language (as reviewed in Weizenbaum 1976). Today, according to some, the
advent of large language models means that computers now have "mastered
language"
([source](https://devm.io/machine-learning/chatgpt-openai-agi)) or at
least can "acquire human-like grammatical language" (Contreras Kallens
et al. 2023).

<figure class="wp-block-image size-large is-resized">
<a
href="https://ideophone.org/files/hristina-satalova-IVvPluBxfxI-unsplash-scaled.jpg"><img
src="https://ideophone.org/files/hristina-satalova-IVvPluBxfxI-unsplash-1024x683.jpg"
class="wp-image-6902" style="width:640px;height:auto" loading="lazy"
decoding="async"
srcset="https://ideophone.org/files/hristina-satalova-IVvPluBxfxI-unsplash-1024x683.jpg 1024w, https://ideophone.org/files/hristina-satalova-IVvPluBxfxI-unsplash-300x200.jpg 300w, https://ideophone.org/files/hristina-satalova-IVvPluBxfxI-unsplash-768x512.jpg 768w, https://ideophone.org/files/hristina-satalova-IVvPluBxfxI-unsplash-1536x1024.jpg 1536w, https://ideophone.org/files/hristina-satalova-IVvPluBxfxI-unsplash-2048x1365.jpg 2048w, https://ideophone.org/files/hristina-satalova-IVvPluBxfxI-unsplash-1600x1067.jpg 1600w"
sizes="(max-width: 1024px) 100vw, 1024px" width="1024"
height="683" /></a>
<figcaption>Photo by <a
href="https://unsplash.com/@hristinash?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Hristina
Šatalova</a> on <a
href="https://unsplash.com/photos/man-standing-front-of-brown-cattle-on-grass-field-IVvPluBxfxI?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Unsplash</a></figcaption>
</figure>

## Terra nullius {#terra-nullius .wp-block-heading}

So we are hesitant to grant non-human animals even the tiniest bit of
linguistic capacities, and yet over-eager to ascribe mastery of language
to computers. Why? One reason is the Cartesian history of linguistics
and AI: a mentalist-rationalist project that sees language as an
abstract symbol system and the mind as an abstract symbol manipulator.
As Weizenbaum (1976) wrote, "we, all of us, have made the world too much
into a computer".

One effect of this is that objectively interesting features of
human-animal interaction end up in a kind of interdisciplinary *terra
nullius*. This despite the fact that there is plenty to study. How do
allospecifics (agents from different species) manage to achieve
interactive contingency? How do different bodies co-construct semiotic
affordances in interaction? How do we manage to assign meaning to the
behaviour of ostensibly non-linguistic others?

Another effect, equally striking, is that our bedazzlement with the
fluid outputs of text generators leads us to overlook the very serious
interactive and
[interpretive](https://doi.org/10.1145/3442188.3445922){target="_blank"
rel="noreferrer noopener"} work it takes for people to interact with
these machines. We are overeager interpreters, always ready to invest
things with meaning.

In this sense, then, pet owners are not all that different from
wide-eyed LLM enthusiasts. And it is one of the jobs of today's language
scientists to take a cold hard look at the interactive and interpretive
processes going on in both cases. For this, essentialist Cartesian
conceptions of 'mind' and 'language' are not all that useful; indeed
they will frequently stand in the way, biasing us to ignore all that is
flexible and co-constructed about our interaction with non-humans and
inviting us to privilege the disembodied, text-bound artefacts of text
generators. To really crack this nut, we will need to enrich the
language sciences with insights from [disparate
disciplines](https://doi.org/10.1098/rstb.2021.0110){target="_blank"
rel="noreferrer noopener"} like ethology, conversation analysis,
semiotics, embodied cognitive science, and interaction studies.

## The Turing test is overrated {#the-turing-test-is-overrated .wp-block-heading}

Just as Descartes made language a criterion of mind, so Turing made it a
test of machine intelligence. Much is made today of the ability of large
language models (LLMs) to generate text that is sufficiently fluid and
interactive to "pass" the Turing test, i.e., to fool some people some of
the time into thinking they're dealing with a person. I think this is a
category mistake. As we write in our paper:

> The Turing test---a closed experimental setup in which a human
> interpreter judges textual output---foregrounds only the tiniest and
> most disembodied sliver of language. Today's large language models
> (next-token predictors that excel at completing text prompts in
> plausible ways) can pass at least some forms of this test. What do we
> learn from this? Whereas some have rushed to the conclusion that this
> means statistical learning may explain the human capacity for
> language, here we take a different view: it is time to rethink the
> disembodied, decontextualized, text-bound conception of language these
> models are founded on.

In short, the Turing test is overrated. One of my favourite slowly-cited
^[2](#footnote_1_6861 "By the way, “slowly-cited” is a term I just made up for interesting work that somehow hasn’t reached escape velocity; this paper has <5 cites as I write, two of which are ours."){#identifier_1_6861
.footnote-link .footnote-identifier-link}^ papers, McIlvenny
([1993](https://doi.org/10.1515/JISYS.1993.3.2-4.119){data-type="link"
data-id="https://doi.org/10.1515/JISYS.1993.3.2-4.119" target="_blank"
rel="noreferrer noopener"}), mentions at least ten ways in which seeing
it a proof of "machine intelligence" or "mastery of language" is less
than useful, if not actively misleading. Needless to say, Turing himself
wasn't so categorical either; his original essay is well worth a careful
read.

## Recalibrating "linguistics as we know it" {#recalibrating-linguistics-as-we-know-it .wp-block-heading}

Anyway, the point of our short essay is to draw attention to this odd
asymmetry and ask what we can learn from this about our conceptions of
language and about our interactions with non-humans, both animals and
computers.

One possibility is that it will not "have any bearing on linguistics as
we know it" (as one skeptical
[reply](http://www.let.rug.nl/zwart/docs/cornips.pdf){target="_blank"
rel="noreferrer noopener"} to [Cornips'
call](https://doi.org/10.1075/avt.00015.cor){target="_blank"
rel="noreferrer noopener"} for an 'animal turn' in linguistics opined
recently). But of course, there is nothing sacred about "linguistics as
we know it" --- indeed, the very notion is question-begging. A key
challenge for any science is how to escape conceptualizations that limit
us to the field "as we know it" and stay open to recalibration. As
semiotician, linguist and networker extraordinaire Victoria Welby wrote
almost a century ago:

> The living world is plastic, and it is life which has mind and
> language whereby to express it. Thus rigid definition (and the results
> obtained by it) must always be secondary, plastic definition primary.
> Language must mainly follow the inexhaustible subtleties of organic
> phenomena.
>
> Welby 1931, reprinted in Welby 1985 (ed. H.W. Schmitz)

We think there's something to be said for an inclusive view of the
language sciences that doesn't exclude animals from consideration by
default, that is mindful of the limitations of equating language with
abstract symbol systems, and that is open to the study of gradient,
semiotically diverse resources for interaction. Have a read:

Rasenberg, Marlou, Azeb Amha, Marjo van Koppen, Emiel van Miltenburg,
Lynn de Rijk, Wyke Stommel, and Mark Dingemanse. 2023. 'Reimagining
Language: Towards a Better Understanding of Language by Including Our
Interactions with Non-Humans'. *Linguistics in the Netherlands* 40:
309--17. doi:
[10.1075/avt.00095.ras](https://doi.org/10.1075/avt.00095.ras).

## References cited {#references-cited .wp-block-heading}

- Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and
  Shmargaret Shmitchell. 2021. 'On the Dangers of Stochastic Parrots:
  Can Language Models Be Too Big?' In Proceedings of the 2021 ACM
  Conference on Fairness, Accountability, and Transparency, 610--23.
  Virtual Event Canada: ACM.
  [doi.org/10.1145/3442188.3445922](https://10.1145/3442188.3445922){target="_blank"
  rel="noreferrer noopener"}
- Cornips, Leonie. 2019. 'The Final Frontier: Non-Human Animals on the
  Linguistic Research Agenda'. *Linguistics in the Netherlands* 36 (1):
  13--19.
  [doi.org/10.1075/avt.00015.cor](https://doi.org/10.1075/avt.00015.cor).
- Heesen, Raphaela, Marlen Fröhlich, Christine Sievers, Marieke
  Woensdregt, and Mark Dingemanse. 2022. 'Coordinating Social Action:  A
  Primer for the Cross-Species Investigation of Communicative Repair'.
  *Philosophical Transactions of the Royal Society B: Biological
  Sciences* 377 (1859): 20210110.
  <https://doi.org/10.1098/rstb.2021.0110>.
- McIlvenny, Paul B. 1993. 'Constructing Societies and Social Machines:
  Stepping Out of the Turing Test Discourse'. *Journal of Intelligent
  Systems* 3 (2--4).
  [doi.org/10.1515/JISYS.1993.3.2-4.119](https://doi.org/10.1515/JISYS.1993.3.2-4.119).
- Turing, A. M. 1950. 'Computing Machinery and Intelligence'. *Mind* LIX
  (236): 433--60.
  [doi.org/10.1093/mind/LIX.236.433](https://doi.org/10.1093/mind/LIX.236.433).
- Weizenbaum, Joseph. 1976. *Computer Power and Human Reason: From
  Judgment to Calculation*. San Francisco: W. H. Freeman.
- Welby, Victoria. 1985. *Significs and Language: The Articulate Form of
  Our Expressive and Interpretive Resources*. Edited by H. Walter
  Schmitz. Foundations of Semiotics, v. 5. Amsterdam ; Philadelphia: J.
  Benjamins Pub. Co.

::: {.footnotes-made-easy-header}
## Footnotes
:::

1.  [Our piece won shared second prize in an essay competition on *Big
    Questions in Linguistics* organised by LOT, the Netherlands Graduate
    School for
    Linguistics.[[\[]{.fme-pre-backlink}[↩](https://ideophone.org/language-between-animals-and-computers/#identifier_0_6861 "Jump back to text"){.footnote-link
    .footnote-back-link
    aria-label="Jump back to text"}[\]]{.fme-pre-backlink}]{.footnote-back-link-wrapper}]{#footnote_0_6861}
2.  [By the way, "slowly-cited" is a term I just made up for interesting
    work that somehow hasn't reached escape velocity; this paper has \<5
    cites as I write, two of which are
    ours.[[\[]{.fme-pre-backlink}[↩](https://ideophone.org/language-between-animals-and-computers/#identifier_1_6861 "Jump back to text"){.footnote-link
    .footnote-back-link
    aria-label="Jump back to text"}[\]]{.fme-pre-backlink}]{.footnote-back-link-wrapper}]{#footnote_1_6861}


Language is what makes us human: one of those things, perhaps
<em>
 the
</em>
one thing, that sets us apart. But there is an interesting asymmetry in our willingness to ascribe linguistic capacities to non-humans: animals tend to be seen as having none, whereas computers are increasingly thought to have mastered language. This asymmetry is the focus of a recent essay I co-authored with a range of people, led by Marlou Rasenberg.

Messaggi di Rogue Scholar

Language between animals and computers