Discover how word embeddings power Worduel's semantic similarity system and make intelligent word guessing possible.

What Are Word Embeddings?

Word embeddings are mathematical representations of words that capture their meaning and relationships. Think of them as coordinates in a high-dimensional space where words with similar meanings are positioned close together. These embeddings are created by training machine learning models on vast amounts of text data, allowing them to learn how words are used in context and what they mean.

In Worduel, word embeddings enable the game to understand that "ocean" and "sea" are semantically similar, even though they share no letters. This understanding comes from analyzing how these words appear in similar contexts across millions of sentences in training data.

How GloVe Works (English)

GloVe (Global Vectors for Word Representation) is the embedding model used for English words in Worduel. GloVe creates word vectors by analyzing global word co-occurrence statistics across a large corpus of text. The key insight is that words appearing in similar contexts should have similar vector representations.

GloVe vectors are typically 300-dimensional, meaning each word is represented by 300 numbers. These numbers encode semantic information: words with similar meanings will have similar values across many dimensions. When we calculate cosine similarity between two word vectors, we're measuring how similar their semantic representations are.

How FastText Works (French)

FastText is the embedding model used for French words in Worduel. Unlike GloVe, FastText represents words as the sum of their character n-grams, which makes it particularly effective for handling morphological variations common in French. This means FastText can understand relationships between words with different endings, prefixes, or conjugations.

For example, FastText can recognize that "manger" (to eat), "mange" (eats), and "mangé" (eaten) are related, even though they have different forms. This morphological awareness makes FastText ideal for French, where word forms vary significantly.

Why Semantic Similarity Matters

Semantic similarity is the foundation of Worduel's gameplay. Traditional word games rely on letter matching or pattern recognition, but semantic similarity allows players to think about meaning and context. This creates a more intellectually stimulating experience that tests your understanding of word relationships rather than just spelling or pattern matching skills.

The semantic approach means that words don't need to share letters to be considered similar. "Ocean" and "sea" rank closely together because they mean similar things, not because they look similar. This opens up entirely new gameplay possibilities and makes Worduel unique among word games.

Technical Explanation for Non-Technical Users

You don't need to understand the technical details to enjoy Worduel, but here's a simple explanation: Imagine every word has a unique "fingerprint" made of 300 numbers. Words with similar meanings have similar fingerprints. When you make a guess, Worduel compares your word's fingerprint to the secret word's fingerprint and tells you how similar they are.

The ranking system sorts all words by how similar their fingerprints are to the secret word. Rank 1 has the most similar fingerprint, rank 2 has the second most similar, and so on. Your goal is to use these similarity clues to find the secret word.

How Embeddings Affect Gameplay

The quality and characteristics of word embeddings directly impact your Worduel experience. High-quality embeddings capture subtle semantic relationships, making the game more intuitive and educational. The embeddings determine which words are in the vocabulary, how words rank relative to each other, and what semantic patterns emerge during gameplay.

Understanding that embeddings create semantic clusters helps you develop better strategies. When you see that multiple words rank similarly, you're observing the embedding space's structure. This knowledge helps you navigate the semantic landscape more effectively and make more informed guesses.