fonemile

Fonemile is a unit of measurement proposed in phonology and computational linguistics to quantify the distance between phonemes. Conceptually, one fonemile represents a single unit of acoustic or perceptual separation within a defined phoneme space. There is no universal scale, and different implementations calibrate fonemiles using various feature representations, such as articulatory feature vectors, phonetic category boundaries, or data-driven embeddings.

Two common approaches describe how fonemiles are computed. A feature-based distance treats phoneme representations as vectors in articulatory or phonetic feature space and computes a distance such as Euclidean distance between phoneme pairs. A perceptual-weighted distance assigns different weights to features according to perceptual confusability studies. In cross-language comparisons, fonemiles aim to provide a common metric for translating inventories and assessing phoneme similarity or likelihood of sound change, though results depend on the chosen representation.

Applications of fonemiles include evaluating models in speech recognition and synthesis, conducting phonological typology studies, and enabling cross-linguistic similarity assessments. In practice, fonemile values are model-dependent and not directly comparable across studies unless the same feature space and calibration are used.

Limitations and debate surround fonemile. Critics note that reducing perceptual and articulatory variation to a single scalar can overlook context dependence, allophony, coarticulation, and sociolinguistic factors. Proponents argue that fonemiles provide a convenient quantitative handle for comparing phoneme systems and for integrating phonological reasoning with computational methods.

See also: phoneme, phonology, articulatory features, phonetic distance, perceptual distance, phoneme embedding, Levenshtein distance.