Better quantifying inter-annotator variability: A step towards citizen science in underwater passive acoustics
Bibliographic record
Abstract
Deployments of underwater passive acoustic recorders have been widely used to study marine biodiversity, especially to detect vocal cetaceans. To process the huge amount of data collected, automatic detection and classification methods are necessary. Recently the development of such methods, which includes training and then testing the models, is mainly based on so-called ground-truth labels, obtained by manual annotation of audio files.However, manual annotation is a difficult and time-consuming process because of the large size of the datasets, the large diversity of the sounds, their unfamiliar representation, the variant quality of the acoustic recordings and the variability in human appreciation.These different factors induce non-negligible differences from one annotator to another, and better quantifying and understanding such differences is capital to make progress in machine learning applications.On this topic, the inter-annotator variability is investigated on three multi-annotator annotation campaigns performed on different marine bioacoustics datasets. Each of them gathered more than 10 annotators with different profiles, from novices to field experts, covering different annotation tasks, different geographical areas and varieties of sound classes. From this multi-annotation, this work enhances the understanding of the inter-annotator variability through the kappa-metrics. In a second part, from a grouping method of annotation based on a majority vote, a drastic reduction of the potential errors in the annotation from novice annotators is observed. This last observation enlightens the possibility of using citizen sciences to overcome the lack of annotation, while maintaining a quality of annotation expected by an expert.
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 | 0.131 | 0.303 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.003 | 0.002 |
| Bibliometrics | 0.005 | 0.005 |
| Science and technology studies | 0.004 | 0.005 |
| Scholarly communication | 0.009 | 0.010 |
| Open science | 0.004 | 0.015 |
| Research integrity | 0.005 | 0.006 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
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".