Large language models unlock the ecology of species interactions

Species interactions shape population dynamics, geographic distributions, evolutionary trajectories, and responses to environmental change. Yet data on these interactions remain scarce across broad spatial, temporal, and taxonomic scales because they are difficult to collect in the field. One promising source of interaction data is citizen science platforms, which contain billions of biodiversity observations, often accompanied by unstructured text comments that may document interactions among organisms. Advance…

By the Truvace deskOctober 15, 20262 min read1 sourcePeer-reviewedTRV-2026-1480· cite & verify

In brief

Researchers built a workflow that applies large language models to the free-text comments attached to biodiversity observations on citizen science platforms. These comments sometimes describe organisms interacting, and the workflow identifies, extracts and categorizes those interactions across multiple languages.

The approach matters because field data on species interactions are hard to collect across wide areas, times and taxonomic groups. Better coverage could help test long-standing ecological questions and track ecological change. The authors say the method still needs validation, expert review and privacy safeguards, and the source reports tests in only two case studies.

Main points

Species interactions affect population dynamics, geographic distributions, evolutionary trajectories and responses to environmental change.
Interaction data are scarce because they are difficult to collect in the field at broad scales.
Citizen science platforms hold billions of biodiversity observations, often with unstructured text comments.
The workflow was tested on bird-bird and plant-pollinator interactions.

The gain

Language models can pull interaction types and participating species out of multilingual citizen science comments with high accuracy, which could widen the coverage of scarce ecological interaction data.

The rundown

The study starts from a data gap: although interactions among species shape many ecological processes, observations are limited in space, time and taxonomic range because collecting them in the field is difficult.

Citizen science platforms offer a possible source, since their observations often include comments that may mention interactions between organisms. Language models make it more feasible to process this unstructured text at scale, including text in multiple languages.

In two case studies, the authors report that the models rapidly extracted both the interaction type and the species involved with high accuracy. They argue the resulting data could improve the ability to characterize, forecast and conserve biodiversity under global change.