Course 02

Represent Your Language Data

Data is the fuel of AI. This course focuses on the critical pre-processing stage, teaching you how to convert human language into a format that machines can process mathematically. Learn about tokenization, embeddings, and responsible data documentation.

Duration: 4 hours 6 Datasets

Key Topics

  • Tokenization: Breaking text into smaller units (tokens)
  • Embeddings: Representing tokens as vectors in high-dimensional space
  • Vector and matrix operations relevant to NLP
  • Data Cards: A framework for transparent dataset documentation

Learning Outcomes

  • Investigate tools for structuring text data
  • Analyze how meaning is represented geometrically in language models
  • Think critically about bias introduced during data preparation
  • Implement tokenizers and create word embeddings
  • Design datasets with focus on transparency and accountability

Practical Projects

  • Hands-on tokenization and embedding generation
  • Ethics exercise: Designing a dataset to mitigate algorithmic bias
  • Train a BPE tokenizer on African languages and compare vocabulary sizes

Prerequisites

  • Completion of Course 01
  • Familiarity with linear algebra (vectors/matrices)

Recommended Datasets

6 datasets are aligned with this course for hands-on exercises and projects.

MAFAND-MT

Machine Translation for 21 African Languages

machine-translation alignment ~5k-30k sentences per language
Bambara Ewe Fon Hausa +5 more

CC-100

Common Crawl Corpus for 100+ Languages

tokenization pretraining Massive (varies by language)
Amharic Hausa Igbo Somali +4 more

FLORES-200

Professional Translations for 200 Languages

translation-evaluation embeddings 3,001 sentences per language
20+ African languages including Dinka, Nuer, Kanuri

Universal Dependencies

Grammar Annotation for African Languages

pos-tagging dependency-parsing Small, high-quality treebanks
Amharic Bambara Coptic Hausa +3 more

OPUS Parallel Corpora

Open Parallel Corpus Collection

machine-translation alignment Millions of sentence pairs
Xhosa Zulu Tswana Sotho +1 more

Kencorpus

Kenyan Languages Corpus: Swahili, Dholuo, Luhya

machine-translation pos-tagging 5.6M words / 177 hours
Swahili Dholuo Luhya

About This Course

This course explores how to represent language data in ways that machines can understand. You’ll learn about vector representations, embeddings, and how to work with multilingual data.

Key Topics

  • Word Embeddings: Learning dense vector representations of words
  • Word2Vec and GloVe: Classic embedding algorithms
  • Multilingual Embeddings: Working with African language representations
  • Contextual Representations: Introduction to modern embedding techniques
  • Embedding Evaluation: Measuring embedding quality

Prerequisites

  • Completion of Course 01 or equivalent knowledge
  • Linear algebra basics (vectors, matrices)
  • Basic neural network concepts