BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.adass.org//adass2026//speaker//RVZF93
BEGIN:VTIMEZONE
TZID:AWST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1;UNTIL=20051231T160000Z
TZNAME:AWST
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
END:STANDARD
BEGIN:STANDARD
DTSTART:20070325T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:AWST
TZOFFSETFROM:+0900
TZOFFSETTO:+0800
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20071028T030000
RRULE:FREQ=YEARLY;BYDAY=4SU;BYMONTH=10;UNTIL=20081025T190000Z
TZNAME:AWDT
TZOFFSETFROM:+0800
TZOFFSETTO:+0900
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-adass2026-G8ABDE@pretalx.adass.org
DTSTART;TZID=AWST:20261105T102400
DTEND;TZID=AWST:20261105T103600
DESCRIPTION:The ALMA Science Archive contains over 455\,000 FITS images spa
 nning a wide range of astronomical sources but currently lacks natural lan
 guage search capabilities. We present a proof-of-concept semantic search s
 ystem operating at the source level: rather than assigning descriptions at
  the project or observing unit level\, we generate one BM25-optimized text
  description per unique astronomical source by aggregating proposal and pu
 blication abstracts from every project that observed that source across th
 e archive. We prompt gpt-4o-mini to extract source type\, physical propert
 ies\, morphological description\, and key scientific findings in a form op
 timized for BM25 retrieval rather than human reading. Of 54\,060 unique so
 urce names identified in the archive\, 10\,758 had at least one associated
  abstract\; 8\,609 received descriptions and results were indexed into Ela
 sticsearch for free-text retrieval. Comparing BM25 on LLM-extracted descri
 ptions against BM25 on raw abstracts across 20 morphological and astrophys
 ical queries\, we find an average overlap of 2.0 out of 5 results\, demons
 trating that the two retrieval methods are complementary rather than redun
 dant. We further implement an image-text bridge using pre-computed visual 
 similarity scores (Stoehr et al. 2025): for each image without a direct so
 urce description\, we assess whether its 30 most visually similar images s
 hare semantically consistent descriptions and assign a description only wh
 en they do\, achieving coverage of 44% to 98% of all 455\,448 images depen
 ding on the consistency threshold applied. This work represents a first st
 ep toward natural language search in the ALMA archive. The pipeline is des
 igned for compatibility with locally-hosted models\, enabling scaling to t
 he full source catalog.
DTSTAMP:20261116T153714Z
LOCATION:Seminar Room 2
SUMMARY:Toward Natural Language Search in the ALMA Science Archive Using LL
 M Source Descriptions - Kiera McCormick\, Adele Plunkett\, Felix Stoehr
URL:https://pretalx.adass.org/adass2026/talk/G8ABDE/
END:VEVENT
END:VCALENDAR
