The Stories We Live By is a crucial resource in the field of ecolinguistics, offering free online courses and materials that delve into the ways language shapes our relationship with the natural world. The course is based on Dr. Arran Stibbe’s book Ecolinguistics: language, ecology and the stories we live by. By examining the stories that permeate our lives, from everyday conversations to global narratives, this resource aids in understanding and transforming the linguistic constructions that impact ecological thinking and actions. The H4rmony Project closely follows the thinsights from “The Stories We Live By” and Dr. Arran Stibbe, who is our co-investigator and advisor. Explore these transformative resources here to deepen your understanding of how language can drive ecological change.

ZEST Letteratura Sostenibile | Tellus, our esteemed media partner, is at the forefront of inspiring ecological thinking among those in positions of power, leadership, and entrepreneurship. ZEST is committed to promote narratives that highlight environmental protection and sustainable living to influence the thoughts and actions of key decision-makers. Their platform, offering content in both Italian and English, delves deeply into ecological themes and serves as a crucial resource for those seeking to understand and advocate for environmental concerns. Collaborating with ZEST allows the H4rmony Project to reach a broader audience, spreading our message of integrating ecological awareness into Generative AI.

The International Ecolinguistics Association is fundamental in our mission, with many of its members actively contributing to the H4rmony Project. This association promotes ecolinguistics and provides an array of resources including the publication of a journal that features ecolinguistic research. Visit their website here to explore a wealth of resources and scholarly articles that inspire and inform our work.

One of the main theoretical foundations of the H4rmony Project is the paper “Ecolinguistics and AI: Integrating Eco-awareness in Natural Language Processing“ by Jorge Vallego, The H4rmony Project Lead Researcher, available on the IEA Journal. This paper discusses the critical need to integrate ecological awareness into AI, particularly within natural language processing systems. It lays the theoretical groundwork for the H4rmony Project by addressing how language shapes human interaction with the environment and the potential of AI to perpetuate or transform these interactions.

The H4rmony Dataset on Hugging Face is a cornerstone of our project. This meticulously curated dataset serves as a vital tool for fine-tuning AI models through reinforcement learning, with each entry specifically categorised by the ecological issues addressed, language, and cognitive structures. It is open source and collaborative, and can be used to train any model directly from code notebooks. There also is a derivative of the dataset pre-formatted to be used via DPO, KTO and ORPO methodologies.

H4rmoniousBreeze is a model trained using the H4rmony dataset, now accessible on Hugging Face. It utilises the insights and principles of ecolinguistics to generate responses that are environmentally conscious and aimed at promoting sustainable behaviours. As a direct application of the H4rmony dataset, this model serves as a showcase for ecological alignment in AI. Even with a small dataset and a one-line system prompt, this model consistently outperforms much larger models in terms of ecological and sustainability alignment, which can be easily tested by trying and comparing prompts.

Theophrastus, is a chat assistant specially developed as part of the H4rmony Project to exemplify the application of ecolinguistics in Large Language Models. Named after the ancient philosopher known for his extensive writings on natural history, Theophrastus is designed to interact with users as an expert in linguistics and natural language processing, who prioritises ecocentrism over anthropocentrism. This AI assistant will always provide an ecolinguistic and ecocentric point of view. Please bear in mind that, as with any small model, factual information provided by Theophrastus ought to be verified prior formal citation.

H4rmonyEval is an instructed model based on Cohere-R+ hosted on Hugging Face. It evaluates responses based on their ecological and sustainability awareness, and their alignment with the H4rmony ecosophy. It assesses question/answers individually or loaded through csv file. It rates the answers on a 0 to 5 scale, in the case of csv files, it gives an average and also produced a new csv file for download with the marks for each entry. This tool is currently used by the H4rmony Project team in their effort to build and validate an ecological and sustainability benchmark, which is one of its main goals.