A Decade o Lexicog aphic Inno a ion
a he Lisbon Academy o Sciences:
s ill om pape o digi al pla o m…
and hen AI came
Ana Salgado
1s In e na ional Con e ence on Lexicology and Lexicog aphy
Ins i u e o Lexicology o he ELTE Hunga ian Resea ch Cen e o Linguis ics
1 Oc obe 2025
Backg ound
Celeb a ing wen y- i e yea s o
dic iona y-making expe ience.
Cu en ole
Coo dina ion o he Lisbon
Academy o Sciences’
lexicog aphic p ojec s.
LexiKon 2025.Ana Salgado
… wha wo ked
… wha ailed
… wha should
come nex
LexiKon 2025.Ana Salgado
Why Do Academy Dic iona ies S ill Ma e ?
Public Good
Unlike comme cial dic iona ies shaped by
ma ke demands, Academy dic iona ies
sa egua d cul u al memo y wi hin a
cen u ies-old adi ion wo h sus aining.
Me hodical App oach
In a wo ld op imised o speed,
unme hodical speed becomes noise. We
p o ide signal: cu a ed e idence, expe
judgemen , and documen ed esul s.
LexiKon 2025.Ana Salgado
LexiKon 2025.Ana Salgado
Lexicog aphic
esou ces
LeXma
DLP
Dic iona y
VOLP
Vocabula y
SD
Specialised dic iona ies
DHB
His o ical-Biog aphical
boleia (PT) = ca ona (BR) [li ]
cabide ( eg.) = c uze a ( eg.) [coa hange ]
impe ial ( eg.) = ino ( eg.) [pin glass o bee ]
zunguei a (AO) = qui andei a (BR) [s ee endo ]
Unique ID
LexiKon 2025.Ana Salgado
LexiKon 2025.Ana Salgado
The Academy’s Long A c
•Founded in 1789, he Lisbon Academy o Sciences placed Po uguese
Po uguese a he hea o i s mission.
•“ ha he composi ion o a dic iona y o he same language should be
should be pa o i s i s wo ks” (ACL, 1793, n. p.)
•The ‘Plan a’ (Plan) o he making o a dic iona y o he Po uguese
Po uguese language was p esen ed a an academic session on 4 July
on 4 July 1780. LexiKon 2025.Ana Salgado
The Academy’s Long A c
1
1793
Dicciona io da Lingoa Po ugueza.
Only eached le e A,
ending wi h “azu a ” ( o b ay). 21976
Dicioná io da Língua Po uguesa.
Again ended wi h le e A – “azu e e”
(a bi d om Timo ).
3
2001
Dicioná io da Língua Po uguesa
Con empo ânea.
Two olumes, 70,000 en ies,
160,000 senses.
LexiKon 2025.Ana Salgado
Ins i u e o Lexicology and Lexicog aphy
o he Po uguese Language (ILLLP)
•In he 1980s, he Academy c ea ed
c ea ed he Ins i u e o Lexicology and
Lexicology and Lexicog aphy o he
he Po uguese Language (ILLLP).
•The ILLLP ca ies he Academy's
lexicog aphic mission: o cul i a e,
cul i a e, de end, and de elop
Po uguese, se ing as an ad iso y
ad iso y body o he Po uguese S a e
S a e on linguis ic ma e s.
LexiKon 2025.Ana Salgado
Losses and Gains
Wha We Los
•Phone ic ansc ip ion couldn' be eco e ed
•E ymology componen had issues wi h
G eek cha ac e s
•La in quan i y ma ks p esen ed p oblems
Wha We Gained
•Au oma ic change epo s
•Sea ch unc ions ac oss headwo ds and de ini ions
•Gi -based backup sys em acking e e y change
LexiKon 2025.Ana Salgado
F om Dead PDF o Li ing Resou ce
T ans o ma ion Comple e
Ha ing an XML e sion was only
he ounda ion...
LexiKon 2025.Ana Salgado
<en y xmlns:m="h p:/Ma h/Ma hML" xmlns: c="h p://www.w3.o g/2007/XMLSchema- e sioning"
xmlns:xsi="h p://www.w3.o g/2001/XMLSchema-ins ance"
xsi:noNamespaceSchemaLoca ion="h p://pe -
ide.di.uminho.p :8080/exis / es /db/schemas/academia-new.xsd">
<me a s a us="impo ed"/>
< e m>/www.w3.o g/1998/
<o h> assou a</o h> <o h> assoi a</o h>
<p on> ɐsˈo ɐ</p on> <p on> ɐsˈoj ɐ</p on> </ e m>
<g oup> <g amG p>n. .</g amG p>
<sense n="1"> <de > U ensílio domés ico o mado po um cabo longo ou cu o ao qual é ixado,
numa das ex emidades, um eixe de olhas de palma, piaçaba, so go, pêlos na u ais ou a i iciais... e
que se e pa a a e o lixo. </de >
<quo e ype="example">O cabo da assou a pa iu-se. Dei ou o a a assou a po que inha os pêlos
gas os.</quo e> < e> < e m>
<o h> assou a mecânica</o h> </ e m>
<sense> <de >mecanismo de esco as olan es, mon ado num eículo, que se acciona
mecanicamen e. </de > </sense> </ e> </sense>
<sense n="2"> <usg ype=" egis e ">Gí .</usg> <de > Saia comp ida; es e de cauda. </de >
</sense>
<sense n="3">[…]
Fou Majo Shi s
Da a Modelling
Adop ed TEI so en ies we e machine-
ac ionable and human
ac ionable
- eadable. Fo ced us o
us o make aci edi o ial knowledge explici .
explici .
Co pus-D i en Edi ing
Mo ed om ‘wha we' e always said’ o ‘wha
o ‘wha he e idence shows’.
Sus ainable In as uc u e
A dic iona y is a wo k low: inges , alida e, edi ,
edi , e iew, publish, moni o . Pipelines and
and checklis s educed ic ion.
New Edi o ial C i e ia
Explici ules o inclusion, labelling, and
upda ing so e e y decision could be jus i ied
jus i ied and eplica ed.
LexiKon 2025.Ana Salgado
The Edi o We Needed:
LeXma
h p://lexma .eu/
<?xml-model h e ="h p://dacl.zb .p /exis / es /db/schemas/academia. ng"
ype="applica ion/xml"
schema ypens="h p:// elaxng.o g/ns/s uc u e/1.0"?><?xml-model
h e ="h ps://exis .dacl.zb .p /exis / es /db/schemas/academia. ng"
ype="applica ion/xml" schema ypens="h p:// elaxng.o g/ns/s uc u e/1.0"?>
<en y xmlns="h p://www. ei-c.o g/ns/1.0" xml:id="DLP-co idiano_1-78de5.xml"
xml:lang="p ">
<dacl:me a xmlns:dacl="h p://dacl.zb .p /anno a ions" s a us="new"/>
< o m>
<o h em="co idiana">co idiano</o h>
<p on>kɔ iˈdjɐnu</p on>
</ o m>
<g amG p>adj.</g amG p>
<sense new="2021-02-04" xml:id="DLP-co idiano_1-78de5.xml-1">
<de > ela i o à co id-19</de >
<x ype="synonymy">
< e ype="en y">co ídico</ e >
</x >
</sense>
<e ym>De <x >< e ype="en y">co id-</ e ></x >+ su ixo <x >< e
ype="en y"
>-iano</ e ></x ></e ym>
</en y>
LexiKon 2025.Ana Salgado
Wha LeXma O e s
S uc u al Edi o
Hides TEI ma kup complexi y while p ese ing
ull XML alidi y. Lexicog aphe s don’ need o
speak XML.
Valida ion
Checks schema con o mi y and in e nal
consis ency.
Fil e ing & S a is ics
Ex ac en y se s by di e en ypes o ma ke s
ma ke s o labels.
Repo s & Ve sioning
Gi -based backups and change logs.
API laye .
LexiKon 2025.Ana Salgado
The Fi s Day Re ela ion
LexiKon 2025.Ana Salgado
LeXma Today
LexiKon 2025.Ana Salgado
Redesigning he Mic os uc u e
Cla i y
De ini ions igh ened, sense s uc u es sha pened, signpos ing imp o ed.
imp o ed. Examples made sho e , mo e cu en , easie o ead.
Inclusi i y
Dic iona y e lec s Po uguese a ie ies ac oss egions and egis e s,
wi h explici labels ancho ed in e idence.
E idence
Labels like a chaic, in o mal, o o ensi e a e g ounded
in co pus da a o documen ed usage, no edi o ial in ui ion.
LexiKon 2025.Ana Salgado
Visual Language Sys em
LexiKon 2025.Ana Salgado
ci a ion
usage example
lexical combina ion
sugges s consul ing he de ini ion o he ixed combina o y in he highligh ed en y
colloca ion
idioma ic exp ession
in o ms abou he use o an ini ial capi al le e
in o ms abou he plu al o m
ma ks he species in senses e e ing o li ing beings
in oduces an encyclopedic no e
in oduces a usage no e
2023: Bi h o he Digi al Edi ion
•The digi al edi ion o he Dicioná io da Língua
Po uguesa was bo n – eely accessible oday
on he Academy's websi e.
•Based on he 2001 edi ion, his digi al e sion
esul ed om collabo a ion be ween na ional
academicians and esea che s, suppo ed by
he Annual Communi y Suppo Fund h ough
FCT.
•The logo ep esen s he le e D; i s se i on
e okes he wo ld o books.
LexiKon 2025.Ana Salgado
h ps://diciona io.acad-ciencias.p /
Lis ening o Use s a Scale
•We can lis en o ou use s a scale.
•Anonymised log iles show us wha people sea ch o and canno ind,
which en ies puzzle hem, and whe e hey abandon he dic iona y.
Iden i y Gaps
Synonym gaps and missing
a ian s
P io i ise Upda es
T igge ed by social o poli ical
deba es
Tes E ec i eness
Whe he labels and c oss-
e e ences help o con use
LexiKon 2025.Ana Salgado
LexiKon 2025.Ana Salgado
Beyond Log Files
Th ee-Le el Classi ica ion Sys em
Le el 0
Wo ds no in he dic iona y (po en ial
neologisms)
Le el 1
Wo ds al eady in he dic iona y
use ul o analysing ends);
Le el 2
Wo ds included bu equi ing
u he edi o ial wo k
Le el 0 candida es a e il e ed –elimina ing ypos, s ange cha ac e s, diminu i es, and ou ine noise – hen e iewed using
e iewed using explici c i e ia o possible inclusion.
LexiKon 2025.Ana Salgado
Use s as Ac i e Collabo a o s
•This case s udy demons a es some hing c ucial:
•use s a e no jus passi e consume s o dic iona ies;
• hey a e ac i e collabo a o s in keeping ou
lexicog aphic esou ces cu en , accu a e, and
esponsi e o eal linguis ic change.
•Thei con ibu ions help us iden i y gaps and eme ging
eme ging usage pa e ns ha migh o he wise escape ou
escape ou a en ion.
LexiKon 2025.Ana Salgado
LexiKon 2025.Ana Salgado
LexiKon 2025.Ana Salgado
The Neology P ocess
Ex ac ion
Ga he candida es om co po a, use
sugges ions, no- esul logs (le el 0), and o he
o he dic iona ies
Con as
Check agains exclusion co pus
o con i m no el y
Selec ion
Apply quan i a i e and quali a i e
c i e ia o inclusion
Inclusion
Comple e edi ion o a new en y
LexiKon 2025.Ana Salgado
No mally, a wo d mus appea 40–50 imes in co po a.
Exclusions: No all equen wo ds a e included. Unnecessa y bo owings, highly echnical e ms, o
socially inapp op ia e o ms may be excluded.
Bu : Some a e wo ds a e included because hey' e cul u ally salien , appea in use sea ches, o
shape public discou se.
LexiKon 2025.Ana Salgado
PORTULAN CLARIN
•PORTULAN CLARIN LVT conso ium, a s a egic ini ia i e
ini ia i e p epa ing Po uguese o digi al and AI e a
e a challenges.
•O iginally o med by he Facul y o Sciences and Facul y o
A s o he Uni e si y o Lisbon, he conso ium was
ecen ly s eng hened by he Lisbon Academy o Sciences.
•I s mission: de elop s a e-o - he-a linguis ic esou ces
esou ces and echnologies o suppo digi al
ans o ma ion.
LexiKon 2025.Ana Salgado
Conso ium P io i ies
LLM In eg a ion
In eg a ing lexical
da abases in o la ge
language models
Rep esen a i e Co po a
Building comp ehensi e
language da ase s
Ad anced Se ices
Fo esea che s,
educa o s, and indus y –
om heu is ic cha bo s o
mul imodal ansc ip ion
•Th ough PORTULAN CLARIN, we aim o ensu e Po uguese is ully ep esen ed in
global AI sys ems, p ese ing cul u al ichness while ein o cing i s echnological ole.
LexiKon 2025.Ana Salgado
Upcoming De elopmen s
1
Heu is ic Assis an
A cha bo buil wi h Re ie al-Augmen ed
Gene a ion (RAG). Use s can upload documen s
and ecei e a ge ed insigh s. 2Mul imodal T ansc ip ion
Ex ending speech- o- ex se ices o ideo and
au oma ically aligning speech wi h ex – i al
o linguis ic esea ch.
3
Seman ic Clus e ing
G ouping ex s, images, and ideos by meaning,
meaning, powe ed by Sen ence T ans o me s
T ans o me s de eloped wi hin he conso ium.
conso ium.
LexiKon 2025.Ana Salgado
h ps://e a is o.ai/
Model wi h 8 billion pa ame e s, specialised o Eu opean
Po uguese, ine- uned o ques ion answe ing, published by he
NLX – Speech and Na u al Language G oup o he Facul y o
Sciences o he Uni e si y o Lisbon (Po ugal) in June 2025.
Key Lessons Lea ned
1
Technology Se es Me hod
Tools mus suppo lexicog aphic
p inciples, no dic a e hem 2Use s A e Pa ne s
Di ec engagemen ans o ms
dic iona ies om s a ic o dynamic
esou ces
3
S anda ds Enable Inno a ion
Open s anda ds c ea e sus ainable,
sus ainable, in e ope able solu ions
solu ions
4E idence D i es Decisions
Co pus da a and use eedback in o m
in o m e e y edi o ial choice
LexiKon 2025.Ana Salgado
Beyond Po ugal
•Models o o he Po uguese-
speaking coun ies
speaking
•P ese ing & p omo ing linguis ic
linguis ic di e si y
•Conc e e s eps:
•Oc obe → aining in Lexicog aphy
Lexicog aphy o he In e na ional
In e na ional Ins i u e o
Po uguese Language
•No embe → Lexical Obse a o y
Obse a o y and ex ension wi hin
wi hin he CPLP
LexiKon 2025.Ana Salgado
LexiKon 2025.Ana Salgado
Shaping he Fu u e Toge he
•Academy dic iona ies ha e g ea es alue when hey combine:
combine:
•Schola ship wi h se ice
•Rigou , inclusi i y, anspa ency
•Digi al pla o ms —and wisely used AI —enable:
•Fas e wo k lows
•Riche , mo e dynamic con en
•G ea e esponsi eness o communi ies
Collabo a ion & S a is ics
✓Ac oss ins i u ions and be ween schola s and use s
use s
✓Collabo a ion wi h di e en esea che s
✓In eg a ion in o PORTULAN CLARIN
✓Daily dialogue wi h use s
✓Academy’s esea ch g an s (suppo o young esea che s)
LexiKon 2025.Ana Salgado
Fundamen al Vocabula y o he Lisbon Academy o Sciences: lexical selec ion, sense alignmen and encoding
Aim: De elop he Vocabulá ioFundamen al o he Lisbon Academy o Sciences as a digi al, co pus-based ool.
Con ex : Designed o Po uguese na i e speake s, especially adul s wi h low li e acy; aligned wi h he Academy dic iona y.
Me hodology: Selec ion o lexical uni s om he iRead4Skills co pus. Sense alignmen wi h he dic iona y. Lexical da a encoded
in TEI Lex-0.
Resul s: Buil a pilo wi h h ee sample en ies. Valida ed me hodology o lexical selec ion, alignmen , and encoding.
Con ibu ion: P o ides a model o co pus-d i en, socially o ien ed lexicog aphy; suppo s educa ion, digi al humani ies, and
inclusion in Po uguese.
De elopmen o a G apheme- o-Phoneme and Syllabi ica ion Sys em o he Fundamen al
Vocabula y
Aim: C ea e a sys em o au oma e syllabi ica ion, s ess ma king, and phone ic ansc ip ion o
Eu opean Po uguese.
Con ex : Applied o Le el 1 (Ve y Easy) en ies o he Vocabulá io Fundamen al o he Academy
Dic iona y.
Me hodology: Regex-based ules implemen ed in Py hon; e ised phonological c i e ia; es ed on 703
lemmas.
Resul s: High success in syllabi ica ion and s ess ma king; phone ic ansc ip ion equi ed manual
checks in excep ional cases.
Con ibu ion: Fi s s ep owa ds scaling o o e 100,000 dic iona y lemmas; signi ican o open-
access compu a ional esou ces in Po uguese.
Thank you
Ob igada
Nisi u ile es quod acimus s ul a es glo ia.
Se não é ú il o que azemos, ã é a gló ia.
Ana Salgado (ACL & UP)
[email protected]
acsalgado@le as.up.p