BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//ItaliaNLP Lab//Officine ItaliaNLP//IT
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Officine ItaliaNLP
X-WR-TIMEZONE:Europe/Rome
BEGIN:VTIMEZONE
TZID:Europe/Rome
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700329T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701025T030000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:2026-07-28-papucci@officine.italianlp.it
DTSTAMP:20261012T084223Z
DTSTART;TZID=Europe/Rome:20260728T150000
DTEND;TZID=Europe/Rome:20260728T160000
SUMMARY:Officine ItaliaNLP — Michele Papucci: Research Stay Report on Int
 erpretability for Controlled Text Generation
DESCRIPTION:Presentation on the Research Stay at the ILLC (University of Am
 sterdam).\n\nhttp://www.italianlp.it/officine/calendario.html#2026-07-28-p
 apucci
LOCATION:Biblioteca ILC\, CNR Area della Ricerca\, Pisa
URL:http://www.italianlp.it/officine/calendario.html#2026-07-28-papucci
STATUS:CONFIRMED
END:VEVENT
BEGIN:VEVENT
UID:2026-09-29-domenichelli@officine.italianlp.it
DTSTAMP:20261012T084223Z
DTSTART;TZID=Europe/Rome:20260929T160000
DTEND;TZID=Europe/Rome:20260929T170000
SUMMARY:Officine ItaliaNLP — Lucia Domenichelli: From global to local: In
 trinsic Dimensionality as a lens on Language Model Representation
DESCRIPTION:Intrinsic dimensionality (ID) estimates how many effective degr
 ees of freedom are needed to describe data embedded in a high-dimensional 
 space. Rather than counting the dimensions of the model’s hidden states\
 , ID aims to recover the lower-dimensional structure on which the represen
 tations may lie. Most applications estimate a single ID value for an entir
 e dataset\, potentially hiding substantial geometric variation across the 
 space. Local and pointwise ID address this limitation by estimating …\n\
 nhttp://www.italianlp.it/officine/calendario.html#2026-09-29-domenichelli
LOCATION:Biblioteca ILC\, CNR Area della Ricerca\, Pisa
URL:http://www.italianlp.it/officine/calendario.html#2026-09-29-domenichell
 i
STATUS:CONFIRMED
END:VEVENT
BEGIN:VEVENT
UID:2026-10-13-domenichelli@officine.italianlp.it
DTSTAMP:20261012T084223Z
DTSTART;TZID=Europe/Rome:20261013T160000
DTEND;TZID=Europe/Rome:20261013T170000
SUMMARY:Officine ItaliaNLP — Lucia Domenichelli: Linguistic Profiling of 
 Transformer Embedding Geometry
DESCRIPTION:Transformer language models embed tokens in high-dimensional sp
 aces\, but whether geometry reflects linguistic structure remains unclear.
  We analyse token representations in BERT and GPT-2\, selected as canonica
 l encoder-only and decoder-only Transformer architectures\, through a ling
 uistically grounded geometric lens. We partition tokens from the UD Englis
 h-EWT treebank by surface and syntactic features (position\, length\, POS\
 , head distance and arity) and examine how their representational geom…\
 n\nhttp://www.italianlp.it/officine/calendario.html#2026-10-13-domenichell
 i
LOCATION:Biblioteca ILC\, CNR Area della Ricerca\, Pisa
URL:http://www.italianlp.it/officine/calendario.html#2026-10-13-domenichell
 i
STATUS:CONFIRMED
END:VEVENT
BEGIN:VEVENT
UID:2026-10-27-dini@officine.italianlp.it
DTSTAMP:20261012T084223Z
DTSTART;TZID=Europe/Rome:20261027T160000
DTEND;TZID=Europe/Rome:20261027T170000
SUMMARY:Officine ItaliaNLP — Luca Dini: Same Data\, Same Architecture\, D
 ifferent Models: Impact of Data Order and Architectural Shape in Small Lan
 guage Models
DESCRIPTION:Training a neural language model involves many design decisions
  about the data\, the architecture\, and the optimisation process. Many of
  these decisions follow established practice: they are passed on from one 
 model generation to the next and refined through experience. Yet the empir
 ical evidence behind these choices is often not publicly documented in a s
 ystematic or reproducible way. Research has addressed many of these questi
 ons\, but often by comparing models that have already been released…\n\n
 http://www.italianlp.it/officine/calendario.html#2026-10-27-dini
LOCATION:Biblioteca ILC\, CNR Area della Ricerca\, Pisa
URL:http://www.italianlp.it/officine/calendario.html#2026-10-27-dini
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR