Spanish Lemma-Based Frequency Bands: Corpus Data and Continuous-Frequency Validation

Spanish Lemma-Based Frequency Bands: Corpus Data and Ambiguity Sensitivity Validation This dataset provides a reproducible framework for constructing and validating lemma-based lexical frequency bands in Spanish. The resource contains 13,871 Spanish lemmas organized into 14 rank-based frequency bands, using frequency estimates derived from SUBTLEX-ESP and wordform–lemma mappings extracted from the AnCora Universal Dependencies treebank. Morphological paradigm size, operationalized as the number of distinct attested surface forms associated with each lemma, serves as an external validation measure. Statistical analyses include Spearman rank correlations, Kruskal–Wallis tests, and negative-binomial regression using log10-transformed lemma frequency. Version 3.0: Ambiguity Sensitivity Analysis This release adds an analysis of ambiguous wordform–lemma mappings. The original allocation procedure assigns the full frequency of an ambiguous wordform to each matching lemma. An alternative procedure divides the frequency equally among candidate lemmas. The analysis identified 1,779 ambiguous matched wordforms. Equal allocation changed the frequency-band assignments of 3,111 lemmas (22.43%). The continuous-frequency Spearman correlation changed from ρ = .385 to ρ = .388, while the negative-binomial incidence-rate ratio changed from 1.467 (95% CI [1.447, 1.487]) to 1.480 (95% CI [1.460, 1.500]). These findings indicate that the aggregate frequency–morphology relationship remains similar under this alternative allocation procedure, although individual lemma rankings and band assignments are sensitive to ambiguity-handling decisions. Reproducibility The release includes the updated Python analysis script, lemma-level ambiguity comparison data, statistical results, and repository documentation. Source code and documentation: https://github.com/drmattpettus/spanish-lemma-frequency-bands The SUBTLEX-ESP and AnCora source resources are required to rerun the full analysis and may need to be obtained separately from their respective distributors.