Daidalos: NER for Literary Studies on Latin and Ancient Greek Texts

Daidalos:
NER o Li e a y S udies on
La in and Ancien G eek Tex s
Nomina Omina: De ec ing and P ese ing Ancien G eek and La in
P ope Names in he Age o A i icial In elligence, Leipzig, 27/06/2024
D . And ea Beye (Humbold -Uni e si ä zu Be lin)
Daidalos
01 NER in Resea ch:
S andalone Me hod
02
NER in Resea ch:
Pa o a Pipeline
03 NER in Teaching
04
Named En i y Recogni ion o Li e a y S udies
on La in and Ancien G eek Tex s
01 | Daidalos
P ojec
In as uc u e
Goals
Why Call a P ojec “Daidalos”?
We …
─de elop an NLP in as uc u e
─ ha will enable esea che s in
Classical Philology and ela ed
disciplines
─ o apply a ious me hods o na u al
language p ocessing
─which a e uncommon in he Ge man
speaking philological communi y.
I was he mos amous
in en o , c a sman, and
builde in an iqui y – o ge my
human ailu es.
daidalos-p ojek .de
Daidalos Pla o m
Menu: NLP-Tools
☑Selec : language, au ho , wo k, ex passage
☑Run
☑Choose be ween NLP me hods NER, POS, Sen imen Analysis
daidalos-p ojek .de

In as uc u e
Mul iple NLP me hods and
co po a, adjus able se ings, pipelines
o li e a y esea ch ques ions,
Iden i y & Access Managemen
Communi y o P ac ice
OA-Publica ion wi h esea ch
andems, lea ning oppo uni ies
(Jupy e No ebooks, H5P), da a bases
on ools and li e a u e, wo kshops
In e p e able AI
T anspa ency & sus ainabili y by
using model ca ds, da a shee s, and
well documen ed e alua ions o
me hods
Goals
02 | NER in Resea ch:
S andalone Me hod
Example
Tagge : Quali y & Applica ions
Challenges & Solu ions
Example
Tagge : Quali y & Applica ions
La in Ancien G eek
Model Name la_co e_web_lg UGARIT/ lai _g c_be _ne
Publica ion Bu ns 2023 Youse e al. 2023
NLP So wa e spaCy Flai NLP
A chi ec u e lo e ec o s
T ansi ion-based Pa se
BERT (T ans o me ) ec o s
Long Sho -Te m Memo y ne wo k
Condi ional Random Field
T aining Da a Caesa , O id,
Pliny (Elde & Younge )
Home , He odo us, A henaeus
Tagse pe sons, loca ions pe sons, loca ions, peoples
Which NER Tagge Should
You Use?
Model Ca ds & Da ashee s o e an
o e iew
How Do You Lea n o Use
NER?
Cu a ed Jupy e No ebooks p o ide
an in oduc ion
Why Should You Lea n o
Use NER?
Unde s anding NER is pa o
imp o ing one‘s own Digi al Li e acies
Teaching is Abou Wha , How, and Why

Model Ca ds …
…accompany he models and p o ide handy in o ma ion
…can be Ma kdown iles wi h addi ional me ada a
…a e essen ial o disco e abili y, ep oducibili y, and sha ing
Bu model ca ds a e di icul …
… o unde s and by a e age esea che s who lack he necessa y digi al li e acies
… o compa e wi h each o he o selec ing he mos sui able agge
Model ca ds should desc ibe …
… he model, i s in ended use, po en ial limi a ions, including biases and e hical
conside a ions, he da a, selec ion o aining and e alua ion, possible
limi a ions, and ecommenda ions, i necessa y
Model Ca d
h ps://anonymous.4open.science/ /se lag-DC3B/documen a ion/model_ca ds/la incy.md
Da ashee s …
…o e ques ion-d i en in o ma ion abou he da ase o a model
…include ques ions on possible sensi i e da a
Bu da ashee s migh con ain oo much in o ma ion ha is no s uc u ed
enough o unexpe ienced use s / esea che s.
Da ashee
h ps://anonymous.4open.science/ /se lag-DC3B/documen a ion/da ashee _la in.md (exce p : only i s pa ag aph)
Jupy e No ebooks as In e ac i e Wo kshee s
─Jupy e No ebooks a e iles ha con ain in e ac i e wo kshee s
─Code can be supplemen ed wi h
a. Tex
b. Colou ed boxes
c. Table o con en s
d. In eg a ion o g aphics o ideos
e. …
─Aim: acquisi ion o new lea ning con en , mo e in-dep h s udy o epe i ion,
easy access o digi al me hods
Bu wo king wi h Jupy e No ebooks is much mo e demanding han i may seem
a i s …

O e iew
Sho me hod
de ini ion
Embedding in
esea ch opic
App oach
Expec ed esul
Le el 1 AI Li e acy
Unde s and he me hod
Fully guided
Use gi en example
Challenges
Using Jupy e
No ebooks
Gene alisa ion
unclea (e.g. any
ex )
Technical
ocabula y (e.g.
lib a y)
Running code and
dealing wi h
po en ial e o
messages
(so wa e
dependencies)
Challenges
Connec
explana ion wi h
code snippe s
Comp ehend
echnical ou pu s
Unde s and and
in e p e esul s
(e.g. esul
accu acy o each
en i y)
Challenges
HTML
Dealing wi h
inco ec esul s
Unde s anding
limi s and
oppo uni ies o
his me hod