Q: Is BUDOVA free to use?

Yes. All text and speech subsets are released under CC-BY 4.0 . Attribution (citation of the dataset card) is the only requirement. Commercial and non-commercial use are both permitted.

Q: Which NER labels does the corpus use?

Eight entity classes: material , tool , process , measurement , structure , safety , regulation , property . Label definitions and span rules are formalised in the annotation guidelines (see /guidelines ).

Q: Is the data available on Hugging Face?

Yes — the subsets live under huggingface.co/budova . The dataset viewer shows live samples from the merged corpus; the API column lets you load each subset with one line of datasets.load_dataset() .

Question 1

Is BUDOVA free to use?

Accepted Answer

Yes. All text and speech subsets are released under CC-BY 4.0. Attribution (citation of the dataset card) is the only requirement. Commercial and non-commercial use are both permitted.

Question 2

How is the data collected?

Accepted Answer

Text comes from construction-domain sources (normative documents, specifications, estimates, field speech) collected via our annotation platform. Each sample passes through ≥3 independent annotators, and the final record is merged by character-level majority voting with Fleiss’ κ reported.

Speech is recorded in-browser by consenting contributors or uploaded as WAV/FLAC files, with speaker IDs anonymised server-side.

Question 3

Which NER labels does the corpus use?

Accepted Answer

Eight entity classes: material, tool, process, measurement, structure, safety, regulation, property. Label definitions and span rules are formalised in the annotation guidelines (see /guidelines).

Question 4

Can I contribute data or annotations?

Accepted Answer

Yes. Apply to join the annotation platform — after approval you can annotate existing tasks, submit new source texts, and add lexicon terms. Every contribution is attributed in the dataset card.

Question 5

How is quality controlled?

Accepted Answer

Every task requires a configurable minimum number of independent annotations (default 3). Merge happens automatically once the threshold is reached; we compute Fleiss’ κ, pairwise span F1 and per-label F1. Entries below a quality cutoff go to admin review.

Question 6

Is the data available on Hugging Face?

Accepted Answer

Yes — the subsets live under huggingface.co/budova. The dataset viewer shows live samples from the merged corpus; the API column lets you load each subset with one line of datasets.load_dataset().

Question 7

How do I cite BUDOVA?

Accepted Answer

Use the BibTeX entry from the Open Data section. The citation covers the dataset as a whole; per-subset citations (speech, text, lexicon) are available in each Hugging Face repository.

Question 8

What about privacy and GDPR?

Accepted Answer

Speech recordings strip identifying metadata at upload; speaker IDs are pseudonymous (SPK-NNN). Contributors can export or delete their personal data at any time through the annotation workspace (GDPR panel).

Questions answered

Join BUDOVA