Multilingual Natural Language Processing — Overview
Course: Multilingual Natural Language Processing (5204MNLP6Y) — 2026/27 Sem. 1, Period 1 Programme: MSc Artificial Intelligence, UvA Credits: 6 EC Instructor: Christof Monz (Informatics Institute) Time zone: Europe/Amsterdam Lectures: Mondays 13:00–14:45 (SP L1.02) and Wednesdays 09:00–10:45 (SP C0.110, room varies), twice weekly through 2026-10-14 Seminar / lab: Wednesdays 11:00–12:45, SP B0.208, Group 4 Exam: Thursday 2026-10-22, 09:00–11:00, Roeterseiland (REC), not Science Park. 2 hours on paper, multiple choice and multiple selection, one two-sided cheat sheet allowed (see MNLP - Exam Analysis)
Prerequisites
- Basic math: linear algebra, calculus, probability
- Basic machine learning: train/dev/test split, linear regression, classification
- Basic implementation experience
Course Content
Provides an overview of NLP problems where multilinguality plays a role:
- Multilingual scenarios: NER for multiple languages, language-independent parsing, cross-lingual classification
- Crosslingual scenarios: Machine translation, crosslingual QA, crosslingual unlearning, crosslingual reasoning
This course is NOT: a comprehensive intro to ML or NLP — it focuses specifically on multilingual/crosslingual aspects.
Assessment
| Component | Weight | Deadline | Notes |
|---|---|---|---|
| Mini Project | TBD | Report, code and slides: Sun 2026-10-11, before 22:00 (extended from Fri 9 Oct 12:00); presentation Mon 12 or Wed 14 Oct | Teams of ~5; picked by 2026-09-04 17:00 |
| Exam | TBD | Thu 2026-10-22, 09:00–11:00 | Rooms: REC B3.05-B3.06-B3.07 / REC B3.08-B3.09. Format below |
The exam is at Roeterseiland, not Science Park
Thursday 2026-10-22, 09:00–11:00, in REC B3.05-B3.06-B3.07 / REC B3.08-B3.09. That is the Roeterseiland campus. Every other MNLP session this period is at Science Park, so do not run the usual commute on autopilot.
Exam format
- 2 hours, on paper.
- Multiple choice and multiple selection questions only.
- Questions are grouped in blocks by topic, but every question is independent of the others.
- One two-sided cheat sheet is allowed, and it can be printed.
- Canvas has a module “Exam” with example questions from previous years of a different course: they show the question type, not this year’s content. Worked through in MNLP - Exam Analysis.
How multiple-selection answers are scored (partial credit, penalties for wrong picks) is not stated anywhere yet.
Mini-project deadline moved
Monz extended the deadline to Sunday 2026-10-11, before 22:00 (announcement of 2026-10-05). The submission form is the assignment “Mini Project Final Report, Slides and Code” in the Mini Project module, which is now visible on Canvas.
Notes
- Monz always requires pass on both assignments and the exam. Realistically, you’ll need at least a 5.5 on each.
- Lectures run twice a week through 14 October, in parallel with the project. The project is not a substitute for attending: the exam is examined off the lectures.
Weekly Schedule
Standing times and rooms, confirmed from MyTimetable, synced 2026-09-16
- Lecture A: Mondays 13:00–14:45, SP L1.02
- Lecture B: Wednesdays 09:00–10:45, SP C0.110, except SP H0.08 on 23 Sep and SP L1.01 on 14 Oct
- Seminar / lab: Wednesdays 11:00–12:45, SP B0.208, Group 4
- Teaching runs through Wed 2026-10-14. Nothing is timetabled after that except the exam.
- Exam: Thursday 2026-10-22, 09:00–11:00, REC B3.05-B3.06-B3.07 / REC B3.08-B3.09 (Roeterseiland)
Week 1 (31 Aug–4 Sep): Overview and Multilinguality
Canvas module “Week 1” holds two decks.
| Topic | Lecturer | Notes | |
|---|---|---|---|
| L1 | Introduction to NLP, applications, course admin | Monz | MNLP-L01 - Overview |
| L2 | Multilinguality and writing systems | Monz | MNLP-L02 - Multilinguality and Writing Systems |
| Key date | Team formation and problem selection by Sep 4 |
Week 2 (7–11 Sep): Morphology and Segmentation
Canvas module “Week 2” holds two decks plus the Zoom recording of 9 September, which was moved online because of the public transport strike.
| Topic | Lecturer | Notes | |
|---|---|---|---|
| L3 | Morphology and word formation | Monz | MNLP-L03 - Morphology and Word Formation |
| L4 | Subword segmentation | Monz | MNLP-L04 - Subword Segmentation |
Read L4 before choosing a mini-project
Subword segmentation is where the tokenizer decisions live, and every mini-project makes them whether or not it thinks about them. L4 also flags an error in the lecturer’s Algorithm 2, which prunes the highest-value tokens as printed.
From week 3, one deck runs across two lectures
Canvas Modules (read 2026-10-05) shows Monz teaching one deck per week and finishing it at the start of the next: week 4 opens with “continuation of static embeddings”, week 5 with “continuation of contextual embeddings”, week 6 with “continuation of crosslingual NLP”. So notes are numbered by deck, and the exact slide where each lecture stopped is not recorded.
Week 3 (14–18 Sep): First Model
| Slot | Lecturer | Notes | |
|---|---|---|---|
| Mon | Mon 14 Sep, 13:00–14:45, SP L1.02 | Monz | MNLP-L05 - Static Embeddings |
| Wed | Wed 16 Sep, 09:00–10:45, SP C0.110 | Monz | MNLP-L05 - Static Embeddings |
| Lab | Wed 16 Sep, 11:00–12:45, SP B0.208 (Group 4) | ||
| Project | Implement first model, evaluate, debug |
Week 4 (21–25 Sep): Refinement
| Slot | Lecturer | Notes | |
|---|---|---|---|
| Mon | Mon 21 Sep, 13:00–14:45, SP L1.02 | Monz | End of MNLP-L05 - Static Embeddings, then MNLP-L06 - Contextual Embeddings |
| Wed | Wed 23 Sep, 09:00–10:45, SP H0.08 | Monz | MNLP-L06 - Contextual Embeddings |
| Lab | Wed 23 Sep, 11:00–12:45, SP B0.208 (Group 4) | ||
| Project | Refine or try alternative model; dropout, layernorm, residual connections |
Week 5 (28 Sep–2 Oct): Error Analysis
| Slot | Lecturer | Notes | |
|---|---|---|---|
| Mon | Mon 28 Sep, 13:00–14:45, SP L1.02 | Monz | End of MNLP-L06 - Contextual Embeddings, then MNLP-L07 - Crosslingual NLP |
| Wed | Wed 30 Sep, 09:00–10:45, SP C0.110 | Monz | MNLP-L07 - Crosslingual NLP |
| Lab | Wed 30 Sep, 11:00–12:45, SP B0.208 (Group 4) | ||
| Project | Second model refinement, error analysis, conclusions |
Week 6 (5–9 Oct): Finalize
| Slot | Lecturer | Notes | |
|---|---|---|---|
| Mon | Mon 5 Oct, 13:00–14:45, SP L1.02 | Monz | Continuation of MNLP-L07 - Crosslingual NLP |
| Wed | Wed 7 Oct, 09:00–10:45, SP C0.110 | Monz | End of MNLP-L07 - Crosslingual NLP: multilingual NMT, mBART, recap (slides 38 to 45 and 54 to 56, added to the deck that morning) |
| Lab | Wed 7 Oct, 11:00–12:45, SP B0.208 (Group 4) | ||
| Due | Sun 11 Oct, before 22:00 (extended from Fri 9 Oct 12:00): 4-page report, code, slides |
Week 7 (12–16 Oct): Presentations, last teaching week
| Slot | Lecturer | Notes | |
|---|---|---|---|
| L13 | Mon 12 Oct, 13:00–14:45, SP L1.02 | Monz | Last Monday slot |
| L14 | Wed 14 Oct, 09:00–10:45, SP L1.01 | Monz | Last session of the course. Room differs from the usual C0.110 |
| Lab | Wed 14 Oct, 11:00–12:45, SP B0.208 (Group 4) | ||
| Project | Team presentations (13 min + 5 min Q&A) | Mon 12 and Wed 14 Oct, in the lecture slots |
Presentations, confirmed 2026-10-05
Monz’s announcement: presentations run in the Mon 12 Oct and Wed 14 Oct lecture slots and lecture rooms. 13 minutes + 5 minutes Q&A, 2 minutes for handover. Each team’s slot is in columns L and M of his sheet. Bring your own laptop, and ideally a spare: he uses his for notes.
Week 8 (19–23 Oct): Exam
| Slot | Notes | |
|---|---|---|
| Exam | Thu 22 Oct, 09:00–11:00 | REC B3.05-B3.06-B3.07 / REC B3.08-B3.09, Roeterseiland campus |
Key Deadlines
| Date | What |
|---|---|
| 2026-09-04 17:00 | Submit team + problem choice (Google Sheet) |
| Sun 2026-10-11, before 22:00 | Submit report (4 pp PDF), slides, code (GitHub link). Extended from Fri 9 Oct 12:00 |
| 12 or 14 Oct | Team presentation, slot in Monz’s sheet |
| Wed 2026-10-14 | Last teaching session |
| Thu 2026-10-22 09:00–11:00 | Exam, Roeterseiland (REC), not Science Park |
Mini Project: What Makes a Good One
- Good succinct description of most relevant research papers
- Good description of data preprocessing/settings
- Good motivation of neural architecture choices
- Thorough evaluation under different settings/architectures; error analysis
- Report focusing on most relevant findings: what works and what doesn’t, why
- MNLP - Mini Project — full project description, schedule, deliverables, and evaluation criteria
Source material
The Files tab returns HTTP 403 for students, so decks come from Modules, through the canvas_modules connector tool (added 2026-10-05), which returns a download URL per file. Module contents as of 2026-10-07:
| Module | Item | Note |
|---|---|---|
| Mini Project | mini-project.pdf, Project Suggestions and Project Requirements pages, team and cluster-credit sheets | MNLP - Mini Project |
| Week 1 | overview.pdf, multilingual-nlp.pdf | MNLP-L01 - Overview, MNLP-L02 - Multilinguality and Writing Systems |
| Week 2 | morphology.pdf, subword.pdf, Zoom recording of 9 Sep (video1951888080.mp4, 180 MB) | MNLP-L03 - Morphology and Word Formation, MNLP-L04 - Subword Segmentation |
| Week 3 | embeddings_static.pdf (71 slides, 404 pages with animation builds) | MNLP-L05 - Static Embeddings |
| Week 4 | embeddings_context.pdf (50 slides, 305 pages) | MNLP-L06 - Contextual Embeddings |
| Week 5 | cross-lingual-nlp.pdf (56 slides, 267 pages; re-uploaded 2026-10-07 with slides 38 to 45 and 54 to 56 added) | MNLP-L07 - Crosslingual NLP |
| Week 6 | Only a “continuation of crosslingual NLP” header | MNLP-L07 - Crosslingual NLP |
| Exam | exam-examples.pdf (2 pages, 2026-10-06): multiple-choice examples from previous years of a different course | MNLP - Exam Analysis |
The decks are 8 to 23 MB each and are not stored in Assets/, matching weeks 1 and 2.
How the lecture notes are written
The lecture notes were rewritten on 2026-10-13 to be read once instead of attending the lectures (vault rules in CLAUDE.md section 2.3). The exam is multiple choice with one-word traps and a cheat sheet, so each note explains how every method works, the exact conditions that matter, and how it differs from the methods it gets confused with. Results are stated as directions and comparisons; the exact figures are in docs/mnlp-cheat-sheet-numbers/, so no number has left the vault.
Length target: about 3,000 words of body per teaching session the lecture took. L01 is a short overview; L02 to L04 took about one session each (2,500 to 4,000 words); L05 about two and a half sessions (6,000 to 7,000); L06 about two (5,000 to 6,000); L07 about three and a half (7,000 to 8,000).
Exam Prep
- MNLP - Exam Analysis: exam format, the example questions from Canvas, and what they mean for revising
Resources
- Mini Project slides (PDF in Assets/)
- Canvas modules: Mini Project, Week 1 to Week 6 (see Source material)