Lang Data And Transparency

Data & Transparency (lang-data-and-transparency)

Language increasingly treats documented, checkable data and reproducible analysis as a mark of serious work: a reader should be able to see the pattern and, where quantitative, re-run the model. But the requirements differ by subfield and evolve, and linguistic data carry ethical obligations to consultants and communities that generic "open data" rhetoric ignores. This skill helps you document, share, and protect your materials appropriately — without over-stating a deposit mandate the journal may not impose.

When to trigger

• Assembling the data/code/annotation to accompany a submission
• Deciding what can and cannot be shared (consultant confidentiality, community agreements, licensed corpora)
• A reader asked for the dataset, the glossed corpus, the sound files, or the analysis script
• Writing a data-availability statement

What "transparent" means at Language (by data type)

Quantitative (experiment / corpus)

• Share the analysis-ready data and the script that reproduces the models, tables, and figures; pin package versions and set seeds. A repository (e.g., OSF) with a readme is the norm.
• If the raw corpus is licensed, share the derived counts + the extraction code so the pipeline is reproducible even when the source text cannot be redistributed.

Elicited / fieldwork

• Provide numbered, Leipzig-glossed examples with sources; where possible, archive recordings and annotations in a language archive (e.g., ELAR, PARADISEC, AILLA, TLA) under access terms the community agreed to.
• Document the elicitation and transcription workflow so another linguist could interpret the data.

Phonetic

• Share measurement scripts (e.g., Praat scripts) and, where consent allows, the sound files or acoustic measurements; state the alignment/measurement settings.

Ethics of linguistic data (do not skip)

• Consent and community agreements govern what may be archived and how; open sharing is not always ethical, and "restricted access" is a legitimate, respectful choice.
• Anonymize speakers where required; do not expose identities via metadata or audio.
• Credit consultants and communities per current best practice and any community protocol.

Calibration (do not over- or under-state, hedged)

Language rewards transparency as craft, but the pack does not assert a specific editor-verified replication mandate — verify the current data-availability policy on the CUP/LSA author pages before claiming any gate. The honest posture: share what you ethically can, document what you cannot, and never present restricted community data as if it were freely open. Illustrative: a variationist study shares the coded token file, the R script, and a codebook on OSF, but keeps the raw interview audio restricted under the community agreement and says so in the data statement — transparent and ethical at once.

Referee/editor conformance check

Execution bridge (StatsPAI / Stata MCP)

Transparency is a property of the path from data to number, and the cheapest way to keep that path honest is not to have retyped anything along it. Full map: execution-with-mcp.

• Run it, then cite it. A number that was computed by a tool call can be recomputed by a reader with the same script; a number that was remembered cannot. If no server is connected, adapt ../../resources/code/ and say in the paper that you did.
• Deposit what reproduces the table, not a spreadsheet of the finished table: script, seed, pinned package versions, and the derived dataset where the licence and the speakers' consent allow it.
• Where consent or endangerment limits deposit (see the ethics section above), archive the analysis code anyway — the code is not the speakers' data, and withholding it is the part reviewers most often read as evasion.

Anti-patterns

• Quantitative results with no shared data or script (reviewers cannot reproduce the model)
• Treating community/consultant data as "open" without consent or agreement
• Exposing speaker identities through metadata, audio, or examples
• Glossed data with no source, so a reader cannot verify the token
• Over-stating a journal replication mandate the live policy does not impose (verify first)

Transparency pass for Language

Treat this skill as an executable review pass, not a prose hint. First lock what evidence underlies each claim; then judge whether the manuscript answers the venue's real reader: linguists who value checkable evidence and who also respect the ethics of working with speakers and communities.

• Do the pass: for each claim, name where the supporting data and code live, the access terms, and any ethical constraint on sharing.
• Return a ledger: give claim / evidence / where-it-lives / sharing-constraint rows so the next agent can act.
• Sibling guard: documentation-heavy work may fit Language Documentation & Conservation; if a sibling owns the contribution, recommend re-routing before polishing.
• Stop condition: do not give submission-ready advice until resources/official-source-map.md has been checked for the current data-availability policy.

Output format

【Data types】experiment / corpus / elicited / phonetic / archival
【Shared】analysis data + script posted (repo/OSF)? [Y/N/NA]
【Glossed examples】numbered, Leipzig, sourced? [Y/N]
【Ethics】consent / community agreement / anonymization handled? [Y/N]
【Data statement】accurate, no over-stated mandate? [Y/N]
【Next】lang-tables-figures

Supplementary resources

• ../../resources/external_tools.md — archives, OSF, glossing and reproducibility tooling
• ../../resources/official-source-map.md — Language data-availability policy live-check