Semantic, Orthographic, and Phonological Biases in Humans' Wordle Gameplay

TL;DR

This study investigates how semantic, orthographic, and phonological cues bias human Wordle gameplay by comparing human guesses to a near-optimal baseline derived from entropy-based heuristics. It introduces five concrete metrics (Levenshtein, semantic via GloVe, Hamming, and rhyme) to quantify biases between consecutive guesses and analyzes their relationship to the near-optimal strategy. The findings show systematic biases toward prior guesses, especially in more unconstrained scenarios, with orthographic and semantic effects largely decoupled from pure letter-change patterns. The work demonstrates that Wordle serves as a tractable environment to study cognitive biases in constrained language tasks, with implications for understanding how priming and lexical processing shape real-time word generation.

Abstract

We show that human players' gameplay in the game of Wordle is influenced by the semantics, orthography, and phonology of the player's previous guesses. We compare actual human players' guesses with near-optimal guesses using NLP techniques. We study human language use in the constrained environment of Wordle, which is situated between natural language use and the artificial word association task

Figures (4)

• Figure 1: $(c_g, c_y, c_b)$: the number of "green" guesses (correct letter in the correct place) is denoted with $c_g$, the number of "yellow" guesses (correct letter in the incorrect place) is denoted with $c_y$, and the number of letter guesses that are incorrect is $c_b$.
• Figure 2: GloVe similarity between guesses vs. Levenshtein Distance
• Figure 3: Distances between neighboring guesses, by amount of constraint on guess
• Figure 4: Distances between neighboring guesses, by amount of constraint on guess