When a Plant Disappears, More Than a Species Is Lost: Nature Maps 5,796 Plants and 156 Languages as a Knowledge Network at Risk of Fracture
Original Chinese title: 一棵植物消失時，失去的不只是物種：Nature 把 5,796 種植物與 156 種語言畫成一張可能斷裂的「知識網」
This feature examines when a plant disappears, more than a species is lost: nature maps 5,796 plants and 156 languages as a knowledge network at risk of fracture while keeping evidence, cultural context and editorial analysis distinct.
莊溪｜認識植物網站｜教育奉獻獎得主
Author of the Plant Recognition website and recipient of an education contribution award.
# When a Plant Disappears, More Than a Species Is Lost: Nature Maps 5,796 Plants and 156 Languages as a Knowledge Network at Risk of Fracture
Author: 莊溪
This feature asks how scientific modelling, digital systems and Indigenous/place-based knowledge and Two-Eyed Seeing can correct one another without flattening their differences. It treats evidence, cultural context and editorial analysis as separate layers and avoids presenting one knowledge system as the final judge of another.
Chapter 1 — 90,536 Records Are Not an Ordinary Plant Encyclopedia
This chapter starts from the documented evidence: a 2026 Nature study combined 700 sources, 90,536 plant-service records, 5,796 native Amazonian plant species and distribution models for 8,429 species. Across three climate scenarios, local Indigenous cultures could on average lose 28–34% of used plant species and 18–23% of related services. The dataset covers 156 Indigenous languages; under the authors’ explicit worst-case language-loss assumption, the knowledge metaweb could shrink by about 26%. The goal is not to rank scientific and Indigenous knowledge, but to identify the different scales, relationships and blind spots each method can reveal.
Chapter 2 — Climate Change Can Alter Whether Plants Remain in Place
Knowledge is more than an extracted data point. Who observed something, where and when it applies, why the observation matters, what risks are attached to error, and who has authority to revise the interpretation can all be part of the knowledge itself.
Chapter 3 — When a Language Breaks, the Edges of a Knowledge Network Can Break
Disagreement between models and place-based observations should be investigated rather than automatically resolved in favour of one side. Spatial resolution, temporal scale, missing variables, sampling design and research purpose all matter.
Chapter 4 — Putting Knowledge into AI Can Preserve It or Extract It Again
Evidence boundaries are essential. The 28–34%, 18–23% and 26% figures are modelled or scenario estimates, not inevitable forecasts. The 26% estimate depends on a worst-case assumption, and knowledge may continue across languages. The study does not include every interaction among deforestation, fire, mining and extreme events. Generative AI should preserve uncertainty rather than invent culturally plausible details that are not supported by sources.
Chapter 5 — Taiwan Can Build a Biocultural Knowledge Graph, Not an Encyclopedia
For Taiwan, the transferable lesson is method rather than cultural content: co-define the question, preserve provenance and permissions, return results for local validation, and keep versions and uncertainty visible.
Yuan Media AI editorial note
A trustworthy Indigenous knowledge system needs more than a large vector database. It should preserve provenance, permission, place, time, community authority, version history and uncertainty. The system must also be able to say “we do not know” when sources do not support a claim.
Evidence boundaries
The 28–34%, 18–23% and 26% figures are modelled or scenario estimates, not inevitable forecasts. The 26% estimate depends on a worst-case assumption, and knowledge may continue across languages. The study does not include every interaction among deforestation, fire, mining and extreme events.
How to read this evidence
Readers should separate direct observations, modelled scenarios, cultural context and editorial analysis. A network or map can make relationships visible, but its lines do not by themselves prove causation or establish that every community will experience the same outcome.
A local knowledge system should begin with permissions, provenance, versioning and community authority rather than with a demand to collect every available name or recording. Searchability is not the same as consent, and a complete dataset is not the same as preserved relationships.
Any Taiwan application should start with a small, reversible question defined with affected people. The process should record who may correct the data, who may restrict access, how uncertainty is displayed and what happens when the proposed interpretation fails.
The case is therefore useful as a method for asking better questions, not as a universal template. Its public value depends on keeping differences among places, institutions, languages and histories visible.
Methods for traceable digital knowledge
For each research fact, preserve a source identifier; for each cultural explanation, preserve community and context fields; and for each editorial inference, mark analysis. Keep prior versions and dates when sources change, and use access levels so public, research and community versions can differ.
A traceable system should keep the original source, the date of review, permission status, local validation and revision history visible to the people who govern the knowledge.
When a source changes or a community corrects an interpretation, publish the change as a new version instead of silently overwriting the earlier record.
Sources
Role-Driven Inquiry
Yuan Media AI | Guided Questions by Role
Select a role first, then review the complete questions from that perspective; clicking a question will display Yuan Media AI’s response in real time.
Select a role
Indigenous plant knowledge holder / cultural worker
Examine the article's evidence, governance boundaries, data rights, and actionable steps from the perspective of a Indigenous plant knowledge holder / cultural worker.
Choose a question
AI use and content-safety disclosure
AI assisted formatting and translation. Source-supported facts, cultural context and editorial analysis are distinguished.