A Decade of Lexicographic Innovation at the Lisbon Academy of Sciences: still from paper to digital platform… and then AI came

A Decade of Lexicographic Innovation at the Lisbon Academy of Sciences: still from paper to digital platform… and then AI came Ana Salgado 1st International Conference on Lexicology and Lexicography Institute for Lexicology of the ELTE Hungarian Research Centre for Linguistics 1 October 2025 Background Celebrating twenty-five years of dictionary-making experience. Current role Coordination of the Lisbon Academy of Sciences’ lexicographic projects. LexiKonf2025.Ana Salgado … what worked … what failed … what should come next LexiKonf2025.Ana Salgado Why Do Academy Dictionaries Still Matter? Public Good Unlike commercial dictionaries shaped by market demands, Academy dictionaries safeguard cultural memory within a centuries-old tradition worth sustaining. Methodical Approach In a world optimised for speed, unmethodical speed becomes noise. We provide signal: curated evidence, expert judgement, and documented results. LexiKonf2025.Ana Salgado LexiKonf2025.Ana Salgado Lexicographic resources LeXmart DLP Dictionary VOLP Vocabulary SD Specialised dictionaries DHB Historical-Biographical boleia (PT) = carona (BR) [lift] cabide (reg.) = cruzeta (reg.) [coat hanger] imperial (reg.) = fino (reg.) [pint glass of beer] zungueira (AO) = quitandeira (BR) [street vendor] Unique ID LexiKonf2025.Ana Salgado LexiKonf2025.Ana Salgado The Academy’s Long Arc •Founded in 1789, the Lisbon Academy of Sciences placed Portuguese Portuguese at the heart of its mission. •“that the composition of a dictionary of the same language should be should be part of its first works” (ACL, 1793, n. p.) •The ‘Planta’ (Plan) for the making of a dictionary of the Portuguese Portuguese language was presented at an academic session on 4 July on 4 July 1780. LexiKonf2025.Ana Salgado The Academy’s Long Arc 1 1793 Diccionario da Lingoa Portugueza. Only reached letter A, ending with “azurrar” ( to bray). 21976 Dicionário da Língua Portuguesa. Again ended with letter A – “azuverte” (a bird from Timor). 3 2001 Dicionário da Língua Portuguesa Contemporânea. Two volumes, 70,000 entries, 160,000 senses. LexiKonf2025.Ana Salgado Institute of Lexicology and Lexicography of the Portuguese Language (ILLLP) •In the 1980s, the Academy created created the Institute of Lexicology and Lexicology and Lexicography of the the Portuguese Language (ILLLP). •The ILLLP carries the Academy's lexicographic mission: to cultivate, cultivate, defend, and develop Portuguese, serving as an advisory advisory body to the Portuguese State State on linguistic matters. LexiKonf2025.Ana Salgado Losses and Gains What We Lost •Phonetic transcription couldn't be recovered •Etymology component had issues with Greek characters •Latin quantity marks presented problems What We Gained •Automatic change reports •Search functions across headwords and definitions •Git-based backup system tracking every change LexiKonf2025.Ana Salgado From Dead PDF to Living Resource Transformation Complete Having an XML version was only the foundation... LexiKonf2025.Ana Salgado <entry xmlns:m="http:/Math/MathML" xmlns:vc="http://www.w3.org/2007/XMLSchema-versioning" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:noNamespaceSchemaLocation="http://perfide.di.uminho.pt:8080/exist/rest/db/schemas/academia-new.xsd"> <meta status="imported"/> <term>/www.w3.org/1998/ <orth>vassoura</orth> <orth>vassoira</orth> <pron>vɐsˈorɐ</pron> <pron>vɐsˈojrɐ</pron> </term> <group> <gramGrp>n. f.</gramGrp> <sense n="1"> <def> Utensílio doméstico formado por um cabo longo ou curto ao qual é fixado, numa das extremidades, um feixe de folhas de palma, piaçaba, sorgo, pêlos naturais ou artificiais... e que serve para varrer o lixo. </def> <quote type="example">O cabo da vassoura partiu-se. Deitou fora a vassoura porque tinha os pêlos gastos.</quote> <re> <term> <orth>vassoura mecânica</orth> </term> <sense> <def>mecanismo de escovas rolantes, montado num veículo, que se acciona mecanicamente. </def> </sense> </re> </sense> <sense n="2"> <usg type="register">Gír.</usg> <def> Saia comprida; veste de cauda. </def> </sense> <sense n="3">[…] Four Major Shifts Data Modelling Adopted TEI so entries were machineactionable and human actionable -readable. Forced us to us to make tacit editorial knowledge explicit. explicit. Corpus-Driven Editing Moved from ‘what we've always said’ to ‘what to ‘what the evidence shows’. Sustainable Infrastructure A dictionary is a workflow: ingest, validate, edit, edit, review, publish, monitor. Pipelines and and checklists reduced friction. New Editorial Criteria Explicit rules for inclusion, labelling, and updating so every decision could be justified justified and replicated. LexiKonf2025.Ana Salgado The Editor We Needed: LeXmart http://lexmart.eu/ <?xml-model href="http://dacl.zbr.pt/exist/rest/db/schemas/academia.rng" type="application/xml" schematypens="http://relaxng.org/ns/structure/1.0"?><?xml-model href="https://exist.dacl.zbr.pt/exist/rest/db/schemas/academia.rng" type="application/xml" schematypens="http://relaxng.org/ns/structure/1.0"?> <entry xmlns="http://www.tei-c.org/ns/1.0" xml:id="DLP-covidiano_1-78de5.xml" xml:lang="pt"> <dacl:meta xmlns:dacl="http://dacl.zbr.pt/annotations" status="new"/> <form> <orth fem="covidiana">covidiano</orth> <pron>kɔviˈdjɐnu</pron> </form> <gramGrp>adj.</gramGrp> <sense new="2021-02-04" xml:id="DLP-covidiano_1-78de5.xml-1"> <def>relativo à covid-19</def> <xr type="synonymy"> <ref type="entry">covídico</ref> </xr> </sense> <etym>De <xr><ref type="entry">covid-</ref></xr>+ sufixo <xr><ref type="entry" >-iano</ref></xr></etym> </entry> LexiKonf2025.Ana Salgado What LeXmart Offers Structural Editor Hides TEI markup complexity while preserving full XML validity. Lexicographers don’t need to speak XML. Validation Checks schema conformity and internal consistency. Filtering & Statistics Extract entry sets by different types of markers markers or labels. Reports & Versioning Git-based backups and change logs. API layer. LexiKonf2025.Ana Salgado The First Day Revelation LexiKonf2025.Ana Salgado LeXmart Today LexiKonf2025.Ana Salgado Redesigning the Microstructure Clarity Definitions tightened, sense structures sharpened, signposting improved. improved. Examples made shorter, more current, easier to read. Inclusivity Dictionary reflects Portuguese varieties across regions and registers, with explicit labels anchored in evidence. Evidence Labels like archaic, informal, or offensive are grounded in corpus data or documented usage, not editorial intuition. LexiKonf2025.Ana Salgado Visual Language System LexiKonf2025.Ana Salgado citation usage example lexical combination suggests consulting the definition of the fixed combinatory in the highlighted entry collocation idiomatic expression informs about the use of an initial capital letter informs about the plural form marks the species in senses referring to living beings introduces an encyclopedic note introduces a usage note 2023: Birth of the Digital Edition •The digital edition of the Dicionário da Língua Portuguesa was born –freely accessible today on the Academy's website. •Based on the 2001 edition, this digital version resulted from collaboration between national academicians and researchers, supported by the Annual Community Support Fund through FCT. •The logo represents the letter D; its serif font evokes the world of books. LexiKonf2025.Ana Salgado https://dicionario.acad-ciencias.pt/ Listening to Users at Scale •We can listen to our users at scale. •Anonymised log files show us what people search for and cannot find, which entries puzzle them, and where they abandon the dictionary. Identify Gaps Synonym gaps and missing variants Prioritise Updates Triggered by social or political debates Test Effectiveness Whether labels and crossreferences help or confuse LexiKonf2025.Ana Salgado LexiKonf2025.Ana Salgado Beyond Log Files Three-Level Classification System Level 0 Words not in the dictionary (potential neologisms) Level 1 Words already in the dictionary useful for analysing trends); Level 2 Words included but requiring further editorial work Level 0 candidates are filtered –eliminating typos, strange characters, diminutives, and routine noise –then reviewed using reviewed using explicit criteria for possible inclusion. LexiKonf2025.Ana Salgado Users as Active Collaborators •This case study demonstrates something crucial: •users are not just passive consumers of dictionaries; •they are active collaborators in keeping our lexicographic resources current, accurate, and responsive to real linguistic change. •Their contributions help us identify gaps and emerging emerging usage patterns that might otherwise escape our escape our attention. LexiKonf2025.Ana Salgado LexiKonf2025.Ana Salgado LexiKonf2025.Ana Salgado The Neology Process Extraction Gather candidates from corpora, user suggestions, no-result logs (level 0), and other other dictionaries Contrast Check against exclusion corpus to confirm novelty Selection Apply quantitative and qualitative criteria for inclusion Inclusion Complete edition of a new entry LexiKonf2025.Ana Salgado Normally, a word must appear 40–50 times in corpora. Exclusions: Not all frequent words are included. Unnecessary borrowings, highly technical terms, or socially inappropriate forms may be excluded. But: Some rare words are included because they're culturally salient, appear in user searches, or shape public discourse. LexiKonf2025.Ana Salgado PORTULAN CLARIN •PORTULAN CLARIN LVT consortium, a strategic initiative initiative preparing Portuguese for digital and AI era era challenges. •Originally formed by the Faculty of Sciences and Faculty of Arts of the University of Lisbon, the consortium was recently strengthened by the Lisbon Academy of Sciences. •Its mission: develop state-of-the-art linguistic resources resources and technologies to support digital transformation. LexiKonf2025.Ana Salgado Consortium Priorities LLM Integration Integrating lexical databases into large language models Representative Corpora Building comprehensive language datasets Advanced Services For researchers, educators, and industry – from heuristic chatbots to multimodal transcription •Through PORTULAN CLARIN, we aim to ensure Portuguese is fully represented in global AI systems, preserving cultural richness while reinforcing its technological role. LexiKonf2025.Ana Salgado Upcoming Developments 1 Heuristic Assistant A chatbot built with Retrieval-Augmented Generation (RAG). Users can upload documents and receive targeted insights. 2Multimodal Transcription Extending speech-to-text services to video and automatically aligning speech with text –vital for linguistic research. 3 Semantic Clustering Grouping texts, images, and videos by meaning, meaning, powered by Sentence Transformers Transformers developed within the consortium. consortium. LexiKonf2025.Ana Salgado https://evaristo.ai/ Model with 8 billion parameters, specialised for European Portuguese, fine-tuned for question answering, published by the NLX – Speech and Natural Language Group of the Faculty of Sciences of the University of Lisbon (Portugal) in June 2025. Key Lessons Learned 1 Technology Serves Method Tools must support lexicographic principles, not dictate them 2Users Are Partners Direct engagement transforms dictionaries from static to dynamic resources 3 Standards Enable Innovation Open standards create sustainable, sustainable, interoperable solutions solutions 4Evidence Drives Decisions Corpus data and user feedback inform inform every editorial choice LexiKonf2025.Ana Salgado Beyond Portugal •Models for other Portuguesespeaking countries speaking •Preserving & promoting linguistic linguistic diversity •Concrete steps: •October → training in Lexicography Lexicography for the International International Institute of Portuguese Language •November → Lexical Observatory Observatory and extension within within the CPLP LexiKonf2025.Ana Salgado LexiKonf2025.Ana Salgado Shaping the Future Together •Academy dictionaries have greatest value when they combine: combine: •Scholarship with service •Rigour, inclusivity, transparency •Digital platforms —and wisely used AI —enable: •Faster workflows •Richer, more dynamic content •Greater responsiveness to communities Collaboration & Statistics ✓Across institutions and between scholars and users users ✓Collaboration with different researchers ✓Integration into PORTULAN CLARIN ✓Daily dialogue with users ✓Academy’s research grants (support to young researchers) LexiKonf2025.Ana Salgado Fundamental Vocabulary of the Lisbon Academy of Sciences: lexical selection, sense alignment and encoding Aim: Develop the VocabulárioFundamental of the Lisbon Academy of Sciences as a digital, corpus-based tool. Context: Designed for Portuguese native speakers, especially adults with low literacy; aligned with the Academy dictionary. Methodology: Selection of lexical units from the iRead4Skills corpus. Sense alignment with the dictionary. Lexical data encoded in TEI Lex-0. Results: Built a pilot with three sample entries. Validated methodology for lexical selection, alignment, and encoding. Contribution: Provides a model for corpus-driven, socially oriented lexicography; supports education, digital humanities, and inclusion in Portuguese. Development of a Grapheme-to-Phoneme and Syllabification System for the Fundamental Vocabulary Aim: Create a system to automate syllabification, stress marking, and phonetic transcription for European Portuguese. Context: Applied to Level 1 (Very Easy) entries of the Vocabulário Fundamental of the Academy Dictionary. Methodology: Regex-based rules implemented in Python; revised phonological criteria; tested on 703 lemmas. Results: High success in syllabification and stress marking; phonetic transcription required manual checks in exceptional cases. Contribution: First step towards scaling to over 100,000 dictionary lemmas; significant for openaccess computational resources in Portuguese. Thank you Obrigada Nisi utile est quod facimus stulta est gloria. Se não é útil o que fazemos, vã é a glória. Ana Salgado (ACL & UP) [email protected] [email protected]