Michał Kornacki, PhD

Course Description

Course Description

This class is designed to familiarise students with the operation and specifics of translator's work using CAT tools and audiovisual translation programmes. Students will learn how to use a selected CAT tool (settings, projects, translation memories, terminology databases, project management) and translate texts of various types and file formats. Verification of translation with CAT tools will be discussed. In addition, students will be introduced to basic audiovisual translation techniques (theory and practice — subtitling and editing software, audio transcription exercises). Through hands-on projects and critical discussions, students will also explore machine translation integration and the emerging role of AI in translation processes, gaining the technical skills and theoretical understanding needed to navigate the rapidly changing landscape of modern translation.

Languages of the course

The languages of the course (incl. written materials on the webpage) are English and Polish. Students are required to have native-like competence in at least one language, and very good command of the other.

Requirements:

• A T T E N D A N C E
• class contribution - participation in all in-class activities, incl. discussion on home and in-class assignments, teamwork, follow-up practice
• graded in-class assignments

Positive final grade

Pass grade is calculated on the basis all graded assignments during the semester (3 tests), in-class activity and attendance.

Grading system used in my courses

The course uses a balanced grading system where all assessments carry equal weight. The final grade is calculated using the average of all scores earned throughout the semester (including any retakes), along with any additional grade modifiers outlined below.

Tests

Final Grade

Grade modifiers

Students may have up to 2 absences per semester without penalty. Any additional absences require a certified excuse. If you exceed the 2 absence limit without a certified excuse, each additional absence will reduce your final grade by 0.5 points. Additionally, any student who misses more than 50% of classes (even with excuses) will be required to repeat the semester.

Retakes

Retakes are available only at the end of the semester. Please note that all grades count toward your final average. For example, if you initially score a 2, then take a retake and score a 5, your average will be 3.5. It's in your best interest to prepare thoroughly for your first attempt.

A grace period of 10 minutes will be allowed for late arrivals. If you arrive more than 10 minutes late, you will need to attend an alternative class session during the same week. Please note that this accommodation cannot become a regular practice.

What is CAT?

What is CAT?

Computer-Assisted Translation (CAT) can be referred to as a set of tools and methodologies used by professional translators and focused on using computers and information systems in the translation process. The main purpose of CAT systems is to accelerate work and improve translation quality.

Currently, many CAT systems are available on the market. Below is an alphabetical list of CAT systems currently available:

• Across Personal Edition
• Atril Déjà Vu X Professional
• CafeTran Espresso
• Crowdin
• Déjà Vu X
• Fluency Translation Suite
• GeoWorkZ - Translation Workspace
• Heartsome Translation Studio
• LogiTerm Pro
• Lokalise
• MadCap Lingo
• MateCat
• memoQ
• memoQ Cloud
• Memsource
• MetaTexis for Word
• MultiTrans Prism Freelance
• OCLanguage
• OmegaT
• Phrase
• SDL Trados Studio
• Smartcat
• Smartling Translation Management System
• Swordfish Translation Editor
• Text United Translation Software
• Transifex
• Transit NXT Professional+
• Verifika - Quality Assurance Tool for Translators
• Wordbee
• Wordfast
• XTM Cloud
• and more

The main operating principle of these systems is to remember translation units (TU), such as sentences, and suggest them as possible translation options when a similar fragment of source text requires translation. As they developed, successive generations of CAT systems were enhanced with the ability to create terminology databases, corpora, or translation memories based on previous translations. CAT systems are currently used by both freelance translators and translation agencies, where project coordinators and reviewers use them to work with translation projects.

EXERCISE

Phrase Rules!

Phrase rules!

Whatever resources you create (project, TM, TB, etc.), always name them starting with your Surname, e.g., Kornacki_Projekt1.

Exercises

Do NOT delete your translation memory.

EXERCISE

Create a new project and use the TMX you created for the previous exercise to translate the following document using Phrase.

EXERCISE

Translate the following document using Phrase. Discuss the final product in-class.

TMs — Creation, Export and Import

TMs - Creation, Export and Import

Download the following ZIP archive and unpack it.

Revision

EXERCISE

STEPS

#1

Create a new project and import the Source.docx document for translation. DO NOT use machine translation.
Translate the first segment on your own.
Report to your teacher before continuing with step #2

#2

Download PL.docx and EN.docx documents.
Align both documents using your CAT tool of choice in order to create Translation Memory based on those two files.
Import the TM into the CAT tool.
Report to your teacher before continuing with step #3

#3

Translate the remaining part of the document.
Report to your teacher before continuing with step #4

#4

Export your translation to DOCX and your TM as a TMX file and present both to your teacher.

TIME LIMIT: 45 minutes

What is a Good Prompt?

THEORY

1. What is a Prompt?

A prompt is an input instruction or query given to a large language model (LLM) to elicit a specific response. It is the primary interface between human intent and machine output. The basic anatomy of any prompt includes four components: instruction, context, constraints, and output format (even when some of these remain implicit).

2. Why Good Prompts Matter

• Output quality: precise prompts yield precise responses; vague prompts yield generic or irrelevant content
• Efficiency: well-constructed prompts reduce the need for multiple iterations and clarification rounds
• Error reduction: unclear prompts increase the risk of hallucinations, misinterpretations, or fabricated information
• User experience: poor prompts lead to frustration and incorrect assumptions about LLM capabilities

3. Characteristics of Effective Prompts

3.1 Specificity and Clarity

State exactly what you want, not approximately. Avoid ambiguity: define technical terms, scope, and boundaries. Use explicit constraints (word counts, exclusion criteria, required elements).

3.2 Context Provision

Supply background the model needs but does not have access to. Include relevant documents, data, or situational parameters.

3.3 Format Specification

Declare the desired structure explicitly (bullet points, essay, table, JSON, dialogue). Specify tone and register (formal academic, conversational, technical).

3.4 Role Assignment (When Useful)

Assigning a persona can shape tone, depth, and perspective (e.g., "You are a linguistics lecturer explaining to undergraduates").

3.5 Iterative Refinement

First attempts are rarely optimal: treat prompting as a drafting process. Document successful prompt patterns for reuse.

4. Quick Reference Template

[ROLE, if useful]: "You are..."
[TASK]: "I need you to..."
[CONTEXT]: "The following background is relevant:..."
[FORMAT]: "Present this as..."
[CONSTRAINTS]: "Avoid..., include..., limit to..."

5. Beyond Individual Prompting: AI Teams and Delegated Workflows

Prompting exists at two distinct levels:

• Level 1: Individual Prompt Writing — The user constructs every instruction themselves, interacting directly with a single LLM.
• Level 2: Delegated Workflows — The user describes a goal to an orchestrator system, which then handles task breakdown, specialist assignment, and execution across multiple AI agents.

Bilingual Export and Review

THEORY

A bilingual export from a CAT (computer-assisted translation) tool is an exported file that contains both the source text and the corresponding translated (target) text together, usually segment-by-segment. It’s used so people who don’t have the CAT tool (clients, reviewers, DTP operators, subject-matter experts) can see and edit translation in context while preserving the alignment between source and target.

What it usually looks like / common formats

• Bilingual XLIFF (XLF/.xliff/.xlf) or tool-specific bilingual XLIFF (round-trip friendly).
• Bilingual DOCX/RTF (two-column Word file or a Word file with source + target segments).
• Bilingual Excel (source and target in adjacent columns).
• Tool-native formats (SDLXLIFF for Trados, memoQ bilingual RTF/DOCX, Memsource bilingual exports, etc.).
• XLIFF is the most reliable for round-tripping and preserving tags/metadata.

When it’s useful (typical use cases)

• Client or SME review: reviewers compare source and translation line-by-line without a CAT tool.
• Proofreading / linguistic QA: easier to spot omissions, mistranslations and alignment issues.
• Desktop publishing (DTP) and layout checks: reviewers work in the original layout or Word and return changes.
• Post-editing MT or legacy translations: align source + target for faster editing or for creating/refreshing a TM.
• Legal / sign-off workflows where reviewers must see the original wording.
• Alignment tasks: creating translation memories from bilingual files.

Typical bilingual-review workflow

1. Translator/CAT user exports a bilingual file (choose format based on how it will be edited).
2. Send file to reviewer with short instructions (what they may change, how to mark comments).
3. Reviewer edits/annotates in the bilingual file and returns it.
4. Translator re-imports the edited bilingual file (or uses the edits as reference), updates the project/TM, performs QA, and finalizes.

Pros and cons - what to watch for

• Pros: convenient for non-CAT reviewers; preserves source/target alignment; supports round-trip imports when using the right format (e.g., XLIFF).
• Cons / risks: if reviewers edit tags, segment IDs or file structure (merge/split segments, remove tags), re-import can fail or lose information. Word Track Changes can complicate re-import. Bilingual DOCX is convenient but less reliable for perfect round-tripping than XLIFF.
• Reviewers should avoid deleting or altering inline tags (placeholders) or changing segmentation unless they understand the consequences.

Best practices / recommendations

• Give reviewers simple instructions: don’t change tags, avoid splitting/merging segments, and indicate how to mark changes (comments or Track Changes) if re-importing.
• Choose format based on purpose: use XLIFF or the tool-native bilingual format if you need clean round-trip import; use bilingual DOCX/Excel for easy client review or layout checks.
• After re-import, run QA checks and update the TM to capture approved changes.
• If reviewers will do heavy editing outside the CAT tool, consider reconciling changes manually in the CAT tool instead of automatic re-import.

EXERCISE

Download the document and translation memory linked below.
Create a new project and TM, import "TM. Towards bilingual review.docx" as a new job and import "TM. Towards bilingual review.tmx" to your TM.
Use the TM to translate the entire document.
Wait for further instructions once finished.

Termbase Creation

THEORY

Termbase Creation

Czasami w nasze (tłumaczy) ręce wpadają dokumenty, które możemy wykorzystać jako podstawę do pamięci tłumaczeniowej (TM) bądź bazy terminologicznej (TB), którą moglibyśmy wykorzystać w narzędziach CAT.
Nie ma problemu, jeżeli obie wersje językowe takiego tekstu/haseł funkcjonują jako dwa oddzielne dokumenty - możemy je wtedy w łatwy sposób ze sobą porównać/sparować (np. w LiveDocs w memoQ) i w ten sposób utworzyć pamięć tłumaczeniową.
Co jednak w sytuacji, gdy jest to prosta lista haseł z tłumaczeniem, mająca np. ponad 500 haseł (patrz poniżej)?

storm troops = oddziały szturmowe
Strategic Air Command = Dowództwo Strategicznych Sił Powietrznych w USA
strategic corporal = "strategiczny kapral" wyszkolony podoficer zdolny do podejmowania odpowiednich decyzji w trakcie tzw. "three block war"
strike force = oddział uderzeniowy
submarine chaser = ścigacz okrętów podwodnych
submarine pen = schron dla okrętów podwodnych

Przetworzenie takiego źródła do bazy terminologicznej wymaga rozdzielenia haseł na dwie wersje językowe i zapisania ich jako dwie oddzielne kolumny w arkuszu kalkulacyjnym, a następnie import do memoQ.
Strategię postępowania ze słownikiem źrdłowym narzuca format zawartych w nim danych.

Prosta dwukolumnowa tabela (najpopularniejsza)

Wygląd: kolumna A = termin źródłowy, kolumna B = termin docelowy (czasem nagłówek „Source / Target”).

invoice | faktura
client  | klient
    

Uwagi przy konwersji: zazwyczaj najłatwiejsza do zaimportowania; sprawdź, czy nie ma ukrytych znaków, czy separator to rzeczywiście jedna para kolumn, usuń puste wiersze i duplikaty.

Arkusz wielokolumnowy z metadanymi

Wygląd: kolumny: Term | Translation | Part of speech | Context | Domain | Note | Status

invoice | faktura | noun | billing document | Finance | use in invoices | approved
    

Uwagi: w memoQ można mapować dodatkowe pola (np. context, note). Trzeba ustalić, które kolumny będą przeniesione do termbase, a które jako notatki. Sprawdź spójność nazewnictwa pól.

Wpisy „w linii” w dokumencie Word / PDF (tekst paragrafowy)

Wygląd: każdy termin w jednym wierszu/paragrafie, często z myślnikiem lub dwukropkiem: „termin — tłumaczenie. [uwaga]”.

invoice — faktura. Use for VAT invoices.
client: klient (także kontrahent)
    

Uwagi: wymaga parsowania (regex lub ręczne). W PDF może być problem z kopiowaniem (łamane wiersze). Upewnij się co do separatora (—, :, =).

Macierz wielojęzyczna (kolumny: termin źródłowy + kilka języków)

Wygląd: Term | PL | DE | FR | ES

invoice | faktura | Rechnung | facture | factura
    

Uwagi: przy imporcie do memoQ musisz rozdzielić na pary językowe lub stworzyć wielojęzyczny termbase; zwróć uwagę na brak tłumaczeń w niektórych kolumnach i na różne warianty terminów.

Format „wolny” / skany / PDF jako obrazy

Wygląd: lista w PDF będącym obrazem, skany, screenshoty albo tabelka w PDF z niejednolitym układem.

Uwagi: wymaga OCR i ręcznej weryfikacji; często pojawiają się błędy rozpoznawania i pęknięte formatowanie. Traktować jako najtrudniejszy przypadek.

Dodatkowe typowe problemy do uwzględnienia przy konwersji

• Wiele tłumaczeń w jednej komórce (oddzielone średnikami/ukośnikami): trzeba wybrać preferowane lub rozdzielić na oddzielne wpisy.
• Notatki/kontekst w tej samej komórce co tłumaczenie — rozdzielić do oddzielnego pola, np. „note”.
• Brak informacji, który język jest źródłowy/docelowy — ustalić przed importem.
• Różne formy zapisu (wielka/mała litera, odmiany) — normalizacja przed importem.
• Znaki specjalne/HTML/formatowanie RTF — usunąć, zachować tylko treść.
• Kodowanie (UTF-8) dla polskich znaków — upewnić się, że plik jest poprawnie zakodowany.
• W PDF i skanach: błędy OCR i złe łamanie wierszy — wymagana weryfikacja ręczna.

EXERCISE

Your Personal AI Research Team — ClassroomAI

THEORY

What this is

ClassroomAI is a ready-to-use AI team that lives on your own computer. It is built around Claude Code and Ollama cloud models, pre-configured so you can start working immediately — no programming knowledge required. You get three specialised agents (STEVE the coordinator, ZACK the researcher, and NOLAN the team builder) who work together to help you research, analyse, and write.

What it is for

This tool is designed for students and researchers who want hands-on experience with AI agentic orchestration. Use it for any personal or academic project where a structured research workflow helps.

How to get started

1. Download the ZIP for your operating system below.
2. Unzip it and open the folder.
3. Double-click the installer (Windows: INSTALL.bat, Mac: INSTALL.command, Chromebook: INSTALL.desktop).
4. Answer three simple questions (folder name, AI model, pre-built config).
5. Create a free Ollama account when prompted and paste your API key.
6. Wait 15–30 minutes while everything installs automatically.
7. Double-click the shortcut created on your desktop to start your team.

What you need

• A computer running Windows 11, macOS, or ChromeOS (with Linux enabled) — ChromeOS installer should work on debian-based Linux distros out-of-the-box
• A stable internet connection
• About 15–30 minutes for the first setup
• A free Ollama account (created during installation)

Components

This installer downloads and installs free, open-source, and/or freely available software. All components are fetched from their official sources only.

Downloads

Once installed, open your shortcut and type: "Hello! Who are you?" Your team will introduce themselves and you can start working.

About Memory and Learning

The agents can remember what they learn and improve over time.* During installation, you will be asked if you want to include a pre-built basic memory configuration. If you choose yes, the agents will retain notes and preferences across sessions, provided you remember to tell the orchestrator to start and close session. If you choose no, you start with a blank slate and can set up memory manually later.

Persistent memory in multi-agent AI systems involves significant technical challenges. For a full discussion of the memory problem, how it affects agent behaviour, and how this system addresses it, see:

Memory in a Multi-Agent AI System: Problems and Solutions

*Memory persistence requires choosing the pre-built configuration during installation. Full technical details available in the linked document above.

Student Workflows

Workflow assignments submitted by students in the 2025/2026 academic year.

Błażejewska — AI Agents Prompt

Bykowska — Agata

Bykowska — Agata (Audio)

Goluch — Roksana

Kaczmarska — Marlena

Kazanecka — Julia

Malinowska — Marta

Wróbel — Wiktoria (Multi-agent Workflow)

Płóciennikowski — Maciej

Wojtal — Alicja (Request for AI)

Rydzewska — Hanna

Siek — Igor

Stefaniak — Martyna

Stefańska — Izabela

Wiśniewska-Emery — Barbara

Student Satisfaction Survey

Take the Survey