Even Simple Habitat Ontologies are Hard to Use
Bibliographic record
Abstract
An essential component in describing, delimiting, and understanding the evolutionary context of a taxon is characterizing the habitats in which the taxon is found. We report on a simple habitat ontology that we have developed, and on our ongoing experience using volunteers to annotate legacy habitat descriptions with terms from the ontology. Our botanical informatics group is building the Canadian Flora Commons, a knowledge platform to aggregate, integrate and facilitate collaboration on information about Canadian plants. Species pages in the Commons are seeded with structured data extracted from authoritative sources such as the Flora of North America (FNA), Flora of British Columbia, etc. In previous TDWG talks (e.g., Sachs et al. 2019), we described our workflow for extracting and structuring morphological data. To understand why habitat descriptions are different and pose a unique set of challenges, consider the following (from Plectocephalus rothrockii in FNA): “Damp soil near streams, roadsides, open pine-oak woodlands and forests”. Here, the single field “habitat” is used to capture environmental conditions, canopy coverage, and taxonomic associations. We also find it often used for geology, climate, etc. Information in the habitat field is often detailed, but it is presented in free text with little editorial guidance, and comparison between treatments within a given flora and among floras is challenging. Environment ontologies that could aid in the standardization of habitat descriptors exist, notably ENVO (ENVironment Ontology; Buttigieg et al. 2016). However, ENVO’s goals have been primarily focused on describing the biomes, environmental features and environmental materials of molecular datasets, resulting in an ontology that thus far does not serve our needs. To our knowledge, no habitat ontology exists that supports species-level use cases (but see the habitat classification scheme developed by the IUCN). To address this, we developed a small and simple habitat ontology by examining over 3000 habitat descriptions across multiple families, and asked “what is the author trying to tell us?”. In our taxonomic treatment authoring tool, being developed as part of another project, we will use this ontology to replace or supplement the single “habitat” field with multiple habitat dimensions (“soil type”, “canopy coverage”, etc.), some with controlled vocabularies (e.g. {open, closed, partial} for canopy coverage). We are also “translating” legacy habitat descriptions into instance data for the ontology. This is a time-consuming process and has the potential to be dependent on interpretations made by the translator. The crowdsourcing experiment described below is aimed at addressing the first issue and quantifying the second. With our centre's support, we recruited a team of volunteers (6–8 at any given time), and taught them how to annotate habitat descriptions with WebProtegé (Horridge et al. 2014). We divided volunteers into two groups, with each group working with the same dataset, so that we could compare results. While a purpose-built habitat ontology offers advantages over existing environment ontologies and a consensus was reached on habitat class definitions (e.g., moisture, elevation, canopy coverage), we discovered that it is difficult to achieve consensus on the application of habitat classes. Between the two groups, shared annotations represented 57% of the total annotations added to terms and phrases and unique annotations represented 43%. This aligns with previous efforts to build a controlled vocabulary for FNA treatments, where differences between term categorizations represented 49% of the effort (Endara et al. (2017)). Amongst classes in our ontology, unique annotations varied between 11% and 76% (see Fig. 1). Our talk will describe our findings, discuss the subjectivity of habitat classes and other difficulties we’ve encountered while building our ontology, and demonstrate the power of a habitat-driven search interface. This interface will live alongside parsed morphological descriptions (see dev.floranorthamerica.org). We invite collaboration towards increasing the robustness and applicability of the ontology.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.013 | 0.060 |
| Meta-epidemiology (narrow) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.001 | 0.003 |
| Bibliometrics | 0.005 | 0.008 |
| Science and technology studies | 0.003 | 0.003 |
| Scholarly communication | 0.009 | 0.031 |
| Open science | 0.004 | 0.010 |
| Research integrity | 0.002 | 0.007 |
| Insufficient payload (model declined to judge) | 0.019 | 0.014 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".