BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-london-2026//talk//JWNWFQ
BEGIN:VTIMEZONE
TZID:GMT
BEGIN:STANDARD
DTSTART:20001029T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:GMT
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T020000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:BST
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-pydata-london-2026-JWNWFQ@pretalx.com
DTSTART;TZID=GMT:20260606T153000
DTEND;TZID=GMT:20260606T161500
DESCRIPTION:Multilingual embeddings are often assumed to place different la
 nguages into a shared semantic space. In practice\, that alignment breaks 
 down in systematic ways.\n\nThis talk explores where multilingual embeddin
 gs work\, where they fail\, and why. Using examples across multiple langua
 ges\, I show how tokenisation\, training data imbalance\, and semantic amb
 iguity shape embedding behaviour in practice\, along with practical diagno
 stics for evaluating multilingual embeddings.
DTSTAMP:20260602T223344Z
LOCATION:Hardwick Hub
SUMMARY:Do Multilingual Embeddings Really Share a Semantic Space? Practical
  Lessons Across Scripts and Languages - Kavit Tolia
URL:https://pretalx.com/pydata-london-2026/talk/JWNWFQ/
END:VEVENT
END:VCALENDAR