Administrator
This area shows the complete overview and to-do tracking. Enter the administrator password.
Wrong password · Mot de passe incorrect
Complete overview
Admin page: all modules and chapters, gated ones included. To be excluded from public builds via build-target config (Phase 0).
Foundations & classical NLP
Foundations of AI & NLP
- The AI landscape what-counts-as-ai · the-nested-fields · generative-ai-and-llms · nlp-and-its-neighbours · four-paradigms
- From rules to machine learning rules · learning-from-data · learning-from-examples · features · classification-and-regression · three-kinds-of-learning · features-in-nlp
- From neural networks to large language models neurons-and-layers · how-a-network-learns · the-black-box · large-language-models
Language as numbers
- Why language is hard what-computers-compute · ambiguity · variation-and-exceptions
- From words to numbers a-number-for-each-word · vectors · one-hot-vectors
- Word embeddings all-words-equally-different · dimensions-with-meaning · the-company-a-word-keeps · embeddings-from-text · embedding-arithmetic · what-a-fixed-vector-misses
- Contextual embeddings one-word-many-vectors · reading-in-both-directions · using-a-pretrained-model
- Attention & transformers the-bottleneck · all-words-at-once · attention · encoders-and-decoders · what-it-changed · impact-on-machine-translation
- Tokenization what-are-tokens · from-text-to-vectors · why-not-words · bpe · bpe-example · try-and-compare-tokenisers
- Consequences of tokenization the-language-bias · spelling-and-counting · counting-in-tokens
Classical NLP
- A tour of NLP tasks what-nlp-does · one-task-up-close · from-tasks-to-systems
- Tokenization (including an example of the rule-based approach) types-and-tokens · segmenting-with-rules · hard-cases · sentences
- Part-of-speech tagging (including an example of machine learning) word-classes-and-tagsets · ambiguity · tagging-with-probabilities · practice-tagging-livre · limits-of-bigrams · tagging-today
- Normalization reducing-variation · lemmatization-and-stemming
- Automatic term extraction (from rules to large language models) 1 to-do what-is-a-term · where-people-disagree · the-task · linguistic-patterns · termhood-and-unithood · the-hybrid-approach · termostat · learning-to-extract · terms-in-two-languages · term-extraction-with-llms · comparing-methods
Corpus analysis
- What is a corpus? what-is-a-corpus · types-of-corpora · what-are-corpora-for
- Working with Sketch Engine discovering-sketch-engine · exploring-existing-corpora · build-a-web-corpus · check-and-tune-your-corpus · the-practice-corpus · wordlist · absolute-and-relative-frequencies · concordances · word-sketch · going-further
Large language models
LLM lifecycle
- LLMs as next-word predictors predicting-the-next-word · what-makes-a-model-large · why-prediction-needs-knowledge · one-token-at-a-time
- Training data how-much-text · where-the-text-comes-from · what-gets-filtered-out · which-languages · training-on-its-own-output
- Pretraining the-training-loop · who-can-afford-to-pretrain · what-a-base-model-does · base-and-foundation-models · what-a-model-does-not-do
- Fine-tuning from-generator-to-assistant · four-stages · learning-from-examples · learning-from-preferences · alphago-and-move-37 · verifiable-rewards
- Alignment what-alignment-is · model-specifications · the-alignment-tax · jailbreaking-and-red-teaming
- Inference the-parameters-are-frozen · the-context-window · choosing-the-token · the-dials · mistakes-compound · tokens-are-thinking-time
- In-context learning steering-not-learning · system-prompts · showing-examples · how-far-context-goes
LLM capabilities & developments
- The jagged frontier the-jagged-frontier · what-they-reliably-fail-at · confidence-without-self-knowledge · abilities-that-appear-with-scale · the-boundary-moves
- Reasoning models thinking-before-answering · what-you-are-shown · where-it-helps-and-where-it-does-not · what-it-costs
- Tools and agents calling-a-tool · tools-do-not-guarantee-accuracy · from-answering-to-acting · computer-use · the-harness · the-risks-of-acting
- Multimodality 1 stub 1 to-do 1 page
Critical issues with LLMs
- Garbage in, garbage out garbage-in-garbage-out · under-other-names
- Hallucinations 1 to-do where-it-comes-from · two-kinds · why-detection-is-hard · mitigating-hallucinations
- Bias where-it-comes-from · bias-in-language-models · amplification · linguistic-hegemony · english-in-other-languages · cultural-defaults · steering-toward-a-culture · correcting-bias · who-decides
- Environmental impact before-the-figures · training-and-inference · what-one-use-costs · efficiency-and-rebound · data-centres · hard-to-measure · putting-it-in-proportion
- Hidden work ghost-work · how-it-is-paid · content-moderation · the-cost-of-moderation
- Power and access market-concentration · dependence · opacity · open-weights · unequal-access
- Jobs and professions freelance-work · salaried-work · forecasts-and-fears
- Skills and learning cognitive-offloading · the-brain-study · skill-decay · assessment · learning-with-a-model
- Privacy and confidentiality 2 to-do what-a-model-gives-back · data-as-business · other-peoples-information · privacy-law
- Copyright and authorship 2 to-do training-without-consent · fair-use-and-fair-dealing · court-cases · opting-out · who-owns-generated-text · output-that-copies · plagiarism-and-disclosure · detecting-generated-text
- The information ecosystem 3 to-do generated-content-online · ai-slop · peer-review · reviewing-rules · the-open-web · model-collapse
- Knowledge and trust knowledge-collapse · telling-real-from-generated · marking-generated-content · what-is-at-stake
Using LLMs
- Choosing a model families-and-tiers · thinking-effort · chat-api-or-local · matching-a-task · your-own-test-set
- LLM leaderboards reading-a-leaderboard · human-preference-rankings · other-languages · translation-benchmarks
- What chat products offer what-a-product-adds · code-and-connectors · deep-research · agents · documents-images-voice
- Prompting what-prompting-is · say-what-you-want · split-the-task · structure-the-prompt · giving-examples · prompt-language · working-with-documents · iterating-or-starting-over · improving-a-prompt · standing-preferences
- Risks and safeguards checking-the-answer · when-prompts-are-attacks · where-your-text-goes
Translation technology
CAT tools
- Translation memories qu-est-ce-qu-une-memoire · memoire-ou-traduction-automatique · le-cycle-complet · segmentation-et-unites · types-de-correspondance · aucune-correspondance · correspondance-partielle · correspondance-a-100-pour-cent · correspondance-en-contexte · alimenter-la-memoire · choisir-ses-memoires · avantages-et-limites
- The CAT-tool ecosystem definition-tao · bases-terminologiques · autres-fonctionnalites · paysage-des-outils · ta-et-gml-dans-la-chaine
- Wordfast: getting started notre-client · organisation-des-fichiers · nommer-ses-fichiers · creer-un-projet · memoire-et-glossaire · les-preferences · premiere-traduction · correspondances-dans-loutil · terminer-le-projet
- Wordfast: tags and machine translation le-document-du-chapitre · balises-dans-lediteur · modes-daffichage · manipuler-les-balises · regles-de-manipulation · apercu · traduction-automatique · pratique-balises
- Wordfast: alignment pourquoi-aligner · autoaligner · aligner-pas-a-pas · exporter-lalignement · lire-le-tmx-produit · corriger-lalignement · importer-la-memoire
- Wordfast: quality assurance and delivery importer-le-projet · analyse-et-rapports · verifications-qa · glossaire-blocage-memoire · revision-dans-wordfast · export-et-nettoyage admin · corrige
Machine translation
- What machine translation is what-translation-asks · four-generations · mt-in-the-workflow
- Before neural networks: rules and statistics rule-based-mt · statistical-mt
- Neural translation and large language models 1 to-do neural-mt · neural-mt-limits · llms-as-translators · prompting-for-translation · what-still-breaks
- The WMT shared task what-a-shared-task-is · what-wmt-found
Speech technology
- Speech technology: types, quality and consequences three-speech-technologies · speech-translation · measuring-asr-quality · speed-and-quality · quality-factors · speech-in-the-profession
- Speech: exercises 🔒 demo-word-dictate · assignment-word-dictate · demo-powerpoint-captions · assignment-powerpoint-captions · demo-heygen · assignment-heygen
Evaluation & transversal skills
Evaluating language technology
- Evaluation foundations 1 to-do systematic-evaluation · evaluation-strategies · the-gold-standard · inter-annotator-agreement · benchmarks · baselines · interpreting-a-score
- Evaluating extraction and classification two-kinds-of-error · precision-recall-f1 · practice-precision-recall · matching-rules · agreement-as-f1 · a-small-evaluation
- Evaluating machine translation many-correct-translations · the-families · bleu · chrf · chrf-demo · practice-bleu-chrf · embedding-metrics · llm-as-judge · mqm-and-esa · human-and-machine · reading-the-scores · mateo
Exchange formats in translation
- Files, tools and markup client-files · why-standards · markup-basics · plain-text-and-lines
- Reading XML what-xml-looks-like · xml-syntax · special-characters-and-namespaces · encoding · what-you-can-change
- TMX, TBX and XLIFF tmx · tbx · xliff · side-by-side
- Interoperability in practice & AI interoperability-in-practice · formats-and-ai
Regular expressions
- Why regular expressions what-a-regex-is · regex-in-translation · regex-and-language-models · testing-your-regex · conventions-and-good-practice
- Describing characters literal-characters-and-the-dot · escaping · character-classes · ranges-and-negation · whitespace-characters · shorthand-classes · accents-and-unicode · practice-characters
- Describing positions line-anchors · word-boundaries · word-boundaries-and-accents · practice-positions
- Repetition and groups counted-repetition · quantifier-shorthands · groups · alternation · greedy-and-lazy
- Review and practice practice-formats · practice-text-patterns · quiz-find-the-regex · quiz-predict-the-matches
Course-gated
Course materials
- TRA2600 · Practical information 🔒 bienvenue · presentation · calendrier · evaluation · ressources-autorisees · examen-mi-session · parcours-regex · attentes · disponibilites · site-et-studium · plagiat
- TRA2600 · Installing Wordfast 🔒 licence · telechargement · installation · activation · premier-lancement · problemes
- TRA2600 · The exchange formats reading 🔒 1 page
- TRA2600 · Assignment briefs 🔒 1 page
- TRA6008 · Practical information 🔒 bienvenue · presentation · projets · calendrier · evaluation · parcours-en-ligne · ou-trouver-quoi
- TRA6008 · Project briefs 🔒 2 to-do 1 page
- Projet 1 · The group's annotations 🔒 accord-du-groupe · comparer-deux-annotations · vote-majoritaire
- Projet 1 · Step by step 🔒 1 page
- Projet 1 · The two systems' scores 🔒 1 page
- Projet 2 · Machine translation 🔒 3 to-do apercu · pas-a-pas · annoter-les-erreurs
Connections
All public chapters, layered by prerequisite depth (top = no prerequisites). Solid lines = builds on; dashed = related. Colours follow the module accents. Each box links to its chapter. "stub"/"to-do" counts above.
Palette
Every colour in the site and what each one is for. All of it is defined in
src/styles/tokens.css; nothing else picks a colour. Module accents are
computed from the part in src/lib/hues.ts. Swatches are painted with the
tokens themselves, so this board cannot drift from the stylesheet.
Structure: what you can act on
The burgundy family, and nothing else uses it.
-
--primarybuttons, radio and checkbox marks -
--primary-textlinks, and hover text on plain surfaces -
--primary-hovera link or button under the cursor -
--primary-ringfocus rings, hovered borders, the language slider -
--primary-softchip fills, focus glow, the open teach term -
--primary-underlinethe rule under a prose link
Blocks: the closed box canon
One green-to-blue idea, 20 dE2000 apart at the closest.
-
--key-pointsKey points, at the top of every page -
--definitionDefinition, the main mention of a term -
--calloutCallout: notes, caveats, practical advice -
--going-furtherGoing further, and Try it -
--evaluationEvaluation, on class pages
Syntax: code blocks and the figures that show code
Shiki writes these into inline styles at build time (astro.config.mjs reads them back out of tokens.css), and the SVG figures carry the same values.
-
--syntax-nameelement and tag names -
--syntax-attributeattribute names -
--syntax-valueattribute values, strings, numbers -
--syntax-punctuationangle brackets, slashes, quotes -
--syntax-keywordkeywords, in languages that have them -
--syntax-commentcomments
States: how an answer stands
More saturated than the boxes, so they read as signals. Always labelled in words too.
-
--state-oka correct answer, a finished exercise -
--state-partiala partly correct answer, an attempt -
--state-wronga wrong answer, an invalid pattern
Hues: the structure palette
Five, one per part plus the near-neutral for what sits outside them. The plain hue is never painted on anything: it is the base every module's shade is mixed from, and the chips on /resources/ use it unshaded. Admin surfaces have no colour of their own: Todo markers, /todo/ chips and the teach view all use the burgundy.
-
--hue-petrolpart: Foundations & classical NLP -
--hue-terracotta/resources/ level: Technical -
--hue-tealpart: Translation technology · /resources/ level: Accessible -
--hue-indigopart: Large language models · /resources/ level: Intermediate -
--hue-plumpart: Evaluation & transversal skills -
--hue-stonesite surfaces belonging to no module (.shell-reference)
Modules: hue from the part, shade from the position in it
-
foundations -
language-as-numbers -
classical-nlp -
corpus-analysis
-
llm-lifecycle -
capabilities -
critical-issues -
using-llms
-
cat-tools -
machine-translation -
speech-technology
-
evaluation -
standards-structured-languages -
regular-expressions
-
course-materials