arXiv Papers (provider)

A provider skill for ovos-common-reading-pipeline-plugin, reading arXiv paper abstracts aloud.

A second real-world test that common-reading works for more than fairy tales - this one introduces a new content_type ("paper"), sourced from a completely different kind of feed (per-category, updated daily).

This skill has no standalone voice interface. It registers no intents and never speaks. It only answers ovos.common_reading.* bus messages, so you also need ovos-common-reading-pipeline-plugin installed and added to your pipeline config for it to be useful at all.

Install

pip install ovos-skill-arxiv-papers ovos-common-reading-pipeline-plugin

Source

Feeds from https://rss.arxiv.org/rss/<category> (RSS 2.0), one subject category at a time - default cs.AI, configurable via this skill's settings (category, e.g. cs.CL, math.QA, q-bio.NC - see arXiv's category taxonomy). Refreshed at most once a day, matching arXiv's own daily feed rebuild schedule.

Only the abstract is read, not the full paper - papers are PDFs, out of scope for TTS narration. The raw feed description includes arXiv's own arXiv:<id> Announce Type: <type> header before the abstract text; this provider strips that off before offering the text.

Translation

Like ovos-skill-ovosblog, this machine-translates titles (for matching and search responses) and abstracts (when reading them aloud) for non-English devices, using whatever ovos-plugin-manager language-translation plugin is configured. Search responses include "machine_translated": true/false for disclosure.

If no translation plugin is available, this provider does not respond to searches at all for a non-English device, rather than silently offering English content the user didn't ask for.

Note for anyone comparing providers: unlike ovos-skill-andersen-tales/ovos-skill-grimm-tales/ ovos-skill-andrew-lang-tales, which check a fixed SUPPORTED_LANGUAGES set at load time and refuse to load at all otherwise, this provider always loads regardless of device language - it can't know in advance whether a translation plugin will be available, so it always registers its bus events and decides per-search instead (see above).

Collection hints

Responds to collection_hint values in the device's own language - e.g. "arxiv"/"archive" (arXiv is commonly pronounced "archive", so that's a likely STT transcription) on English, "videnskabelige artikler" on Danish, "artículos científicos" on Spanish - matched fuzzily against that language's own alias list (see locale/<lang>/collection.voc). Like ovos-skill-ovosblog, this provider works on any device language, so aliases fall back to English for any language we haven't bothered translating - see ovos-common-reading-pipeline-plugin#26. The collection name itself ("arXiv") stays untranslated - it's a proper noun.

Content type

Responds to content_type hints of "paper", "abstract", "research", or "study". Ignores everything else (e.g. "story", "article").

"Surprise me"

A search with no specific phrase but a matching collection_hint (e.g. "read me a paper from arXiv") returns the most recent paper in the configured category, by pubDate.

Terms of use

arXiv explicitly provides these per-category RSS feeds for this kind of use (see arXiv's RSS documentation) - review the Terms of Use for arXiv APIs before adjusting the request frequency below the daily cache TTL here.

Credits

Content and feed from arxiv.org, an open-access e-print repository operated by Cornell University.

Category

Science

Tags

#science #research #arxiv #papers #provider