You Shall Know a Forest

In his 1957 paper Studies in Linguistic Analysis, John Rupert Firth introduced the assertion that "you shall know a word by the company it keeps," a statement that emphasises the significance of context in understanding word meaning. Firth's insight into collocational meaning highlighted the importance of studying words in relation to their habitual co-occurrences. This foundational idea paved the way for the development of large language models (LLMs), which rely on vast datasets to learn patterns of word usage within context. The same phrase invites us to consider the limitations of AI systems that create apparent meaning by reducing sentences to a collection of words that have kept ‘company’ before.

However, as an increasing use of LLMs through software provided by OpenAI and Google becomes common place, how does the computational collocation of words shape how we know natural environments such as forests?

In recent years, as cinema has socialised the idea of AI, it has often turned to the forests to express the complexity of computational systems. Research and development
laboratories are set in coniferous forests (Ex Machina, 2014, Devs, 2020), suggesting fractal geometries as a reference to mathematical systems. The scale of forests are used as environments in which a human protagonist becomes exposed to unfamiliar climates (Her, 2013), and becomes lost without the aid of instrumentation (Ex, Machina 2014).

A short movie “You shall know a forest” was filmed on the traditional countries of Taungurung and Wurundjeri peoples, and Gadigal of the Eora Nation, to connect the representational construction of language through AI tools, with landscapes that become extensions of the ‘wild city’ (Steele 2022). The Firth quote is adapted according to each scene to extend the assertion that…

You shall know a word by the company it keeps
You shall know a city by the company it keeps
You shall know a mall by the company it keeps
You shall know a field by the company it keeps
You shall know a meadow by the company it keeps
You shall know a parkland by the company it keeps
You shall know a woodland by the company it keeps
You shall know a forest by the company it keeps

The movie asks the viewer to consider how natural systems such as forests become part of a way of knowing the world that is increasingly becoming data-driven. Recovering Cronon who reminded us of the production of ‘wildness’ as an idealised notion of nature (1995), Heise extends the challenge for the environmental humanities to study “the integration of digital tools and methods with older humanistic procedures”
(Heise 2017). Written before the release of common tools such as ChatGPT, Heise’s demand for critical enquiry is never more urgent, as the ingestion of words taken from journals, books and libraries creates a universal business for the representation of nature through LLMs.

The “You shall know a forest” movie is connected to a demonstration of the reductionist tendency of LLMs. The Story Exhaustion Generator 2 is a ‘hack’ of ChatGPT in which the reader is encouraged to train the system to not use particular words in a paragraph by the sociologist Avery Gordon. But as you (I hope) will see, that after 4 or 5 uses of other words that have ‘kept company’ the algorithm fails and reverts to the original words demonstrating the limits of its intelligence.

As a small assemblage of practice and writing, I am interested how LLMs perpetuate digital colonialism by ingesting Indigenous knowledge without consent, often commercialising and misrepresenting it in ways that mirror historical patterns of exploitation. These models are trained on vast datasets that may include Indigenous cultural material, leading to misrepresentation, erasure, or distortion of these traditions, as they often don't align with the frameworks LLMs are designed to process. Additionally, the lack of data sovereignty for Indigenous communities means they have little control over how their knowledge is used, reinforcing power imbalances and contributing to the homogenisation of diverse cultural perspectives. However, I would like to note that having moved to Naarm / Melbourne in January 2024, I have only just begun my journey to understand first peoples experiences, let alone concepts of data sovereignty in the context of data-driven economies.


References:
Cronon, W. (1995) The Trouble with Wilderness; or, Getting Back to the Wrong Nature. In Cronon, W. ed., Uncommon Ground: Rethinking the Human Place in Nature. New York:
W. W. Norton & Co.
Firth, J. R. (1957) Studies in Linguistic Analysis. Wiley-Blackwell.
Heise, U. K (2017) Introduction, in Heise, H. K, Christensen, J., Niemann, M. (eds) The Routledge Companion to the Environmental Humanities. Routledge.
Steele, W. (2020) Planning wild cities: human–nature relationships in the urban age. Routledge.