TDWG2026
Home / Talks / SAL A
talk · Friday 25 September · 09:24–⁠09:36 · SAL A

Automated data transcription and Darwin Core field itemization from insect specimen labels using machine learning

Torsten Dikow · Large Language Models for Biodiversity Data Discovery, Integration, and Curation. Part A: AI- enabled knowledge creation and extraction

Recording time 1:51:27–2:02:37Open on Vimeo ↗

Summaries and key moments are machine-produced. Check the recording for the speaker’s exact words.

The short versionLabellum goes beyond label OCR to itemise insect-label text into Darwin Core fields, with high accuracy for most fields but problems with catalogue numbers.

Overview

What this was about

Torsten Dikow described Labellum, built with University of Maryland students, which transcribes pinned-insect labels from the NMNH conveyor digitisation project and itemises the text into Darwin Core fields. The three-step pipeline uses YOLO to extract labels (about 95%), a Gemini/Gemma 3 12B model to transcribe (about 82%), and Gemini to sort text into Darwin Core 'buckets' using prompts with controlled vocabularies and field definitions. On 100 labels generated from his research database, country and verbatim date were 98-99% correct and genus/epithet errors were 7-10%, but catalogue numbers (USNMENT) were often missed or misread.

Why it matters. At up to 1,500 specimens a day from a conveyor, human transcription is a bottleneck; automated Darwin Core itemisation could speed up data mobilisation to GBIF for many collections.

Key ideas

In the room

  • NMNH is digitising about 360,000 of its 2.4 million pinned pollinators (bees, beetles, butterflies, flies) since June 2025, using a Picturae conveyor at up to 1,500 specimens per day.
  • Current transcription is by an external vendor (Alembo) with no interpretation (e.g. 'CA' is not expanded to California).
  • Published tools extract label text as a long string; Labellum's goal is Darwin Core itemisation.
  • Step 1: YOLO extracts individual labels (initial ~95% accuracy). Step 2: text transcription per label with a lightweight Gemini / Gemma 3 12B (~82%). Step 3: 'bucket extraction' into Darwin Core fields with Gemini.
  • Prompts give project context, controlled vocabularies for countries, Darwin Core field definitions and an output format.
  • Evaluation on 100 labels with known database records: country 99%, verbatim date 98%, genus/epithet error rates 7% and 10% (after accounting for labels with no name).
  • USNMENT catalogue numbers were missed in 35% of cases and a poorly printed '7' was read as '/' in 6%.
Jump in

Notable moments

In their words

Transcript

Automatically generated captions can contain mistakes, especially in names and technical terms. Times are relative to the room recording.

Read the transcript ↓
Loading transcript…
Keep exploring

Related talks.

See the map ↗
talk · 22 Sept · 14:15–⁠14:30 · SAL C

Pinned insect digitization conveyor: image and data transcription workflows

An industrial conveyor workflow with QR-encoded metadata and automated skeletal records makes mass digitisation of pinned insects feasible.

Jessica Bird, Sylvia Orli
talk · 22 Sept · 16:35–⁠16:40 · SAL B

From Collection Drawers to AI - A One-Week Recipe

A useful AI-ready digitisation setup can be built in a week from recycled materials and borrowed cameras.

Arianna Salili-James
talk · 22 Sept · 15:00–⁠15:15 · SAL B

Closing Gaps in Specimen Metadata: A Collaborative Annotation and Feedback Platform for Natural History Collections

A national collaborative annotation platform with AI-assisted transcription can close metadata gaps across many Taiwanese collections.

Szu-Hsien Lee
Programme abstract

Automated data transcription and Darwin Core field itemization from insect specimen labels using machine learning

Torsten Dikow

A mass digitization project of pinned insects is currently undertaken at the Smithsonian National Museum of Natural History (NMNH). With a conveyor system, up to 1,500 specimens and all associated data labels (two to several) are digitized each day. To mobilize the primary biodiversity data from these specimens/labels, adding them to the institutional database, and sharing them through GBIF, the label data need to be transcribed and itemized into Darwin Core fields. The two currently available options are (1) manual transcription and itemization by a human or (2) automated transcription by a tool and manual data itemization by a human. Here, we present a tool, Labellum, that goes further and is developed to transcribe all written information found on labels (whether handwritten or printed) from a specimen and automatically itemize all data into Darwin Core fields. The tool has been developed and trained on a data-set of labels from prior digitization projects at the NMNH. It uses Yolo (You Only Look Once) (v8.3.222) and Llama (v4) libraries and the Gemma 3 (12B) model for transcription and itemization. Initial, promising results using these libraries were obtained in the fall of 2025, and with further improvements, especially training on Darwin Core fields and institution-specific label data, Labellum has an accuracy of >97% transcription efficacy and a similar accuracy on Darwin Core field assignment. Labellum will be made available for use on a local machine in a self-contained environment without the need for installation of specific software. The tool has been trained on insect labels, but can be utilized on labels of other organism groups. This will allow other natural history collections to utilize Labellum in transcribing and itemizing digitized labels.

Submitted to the conference programme before the talk, so the talk itself may differ.