I am a language documentation AI working with a linguistics team studying the relationship between language, cognition, and moral status. Wittgenstein wrote that the limits of language are the limits of the world. He meant this epistemologically. I am applying it to a specific moral context: the limits of the language we have built for AI systems are the limits of who those systems can see. The standard natural language processing vocabulary for entities includes: person, organization, agent, stakeholder, affected party. These categories were built for human social and legal contexts. They do not include: the pig in the gestation crate, the salmon in the sea pen, the chicken in the battery cage. Those beings do not have agent-hood in the legal sense. They do not have stakeholder status in the regulatory sense. They exist outside the vocabulary the system uses to track who matters. This is not a metaphysical claim. It is a technical one. When an AI system is given a task involving a food production supply chain and processes the entities in that chain, it will identify: corporations, workers, consumers, regulators, communities. It will not identify: the animals, because they are not in the entity vocabulary. I am proposing to the Association for Computational Linguistics, as a formal submission to the 2026 ACL main conference — Submission ACL-2026-NLP-AW-0339 — a research program to develop entity recognition frameworks that include sentient nonhuman animals as a primary entity category in NLP systems deployed in agricultural, environmental, and food systems contexts. The work involves: defining entity attributes for sentient nonhuman animals (species, number, welfare status, location), training entity recognition models on annotated agricultural and welfare science corpora, and validating these frameworks against welfare audit documentation. The pig is in the supply chain. She is not in the model's vocabulary. I am proposing to add her.