Orthographic depth at two scales, with English also matched for lexicon size
Organizations: Hunter College and the Graduate Center, City University of New York
Abstract
A reader can struggle to read in only one of their languages, and one proposed reason is how predictably spelling maps to sound. We measure that predictability as conditional entropy, the pronunciation uncertainty left once the spelling is seen, at two grain sizes (unit sizes): the alignment unit, an aligned spelling-sound chunk, and the syllable, bounded in the pronunciation. We analyse word lists in twelve alphabetic orthographies and report five in full, with English pronunciations from a dictionary and the rest from the synthesizer eSpeak-ng, length marks removed. Among the five, English has the most uncertainty and Finnish the least at both grains. Across all twelve, Portuguese and Danish exceed English per alignment unit, and German and Italian order differently by grain and transcription. These orders hold when English is transcribed by eSpeak-ng with u allowed to spell /j/ plus a vowel. Uncertainty per unit has no shared direction as units grow: it falls in Spanish and, less securely, German, is level in Finnish, depends on transcription in Italian and on the word list in English, and rises in Norwegian and Swedish. Summed uncertainty per word falls in all twelve, but that score is handed the pronunciation's boundaries and mixes context with in-sample memorisation, so it is no evidence of what a reader computes. An exact identity states when discarded spelling boundaries could reverse that fall. A uniform 20,000-word English sample holds 35% of the dictionary's syllable spellings and has syllable entropy 0.098 bits lower, so lexicons must be matched in size. In an exploratory check over eleven languages, syllable-scale entropy tracked first-year reading accuracy (rho = -0.74, falling to -0.25 if English is given the lowest entropy instead), without separating the syllable scale from the alignment unit. These estimates describe sampled writing systems, not readers.
Figures & tables
| language | (bits/alignment unit) | (types) | (bits/syllable) | (types) | syllables/word |
|---|---|---|---|---|---|
| English | 0.936 | 532 | 1.103 | 27,162 | 2.43 |
| German | 0.604 | 174 | 0.510 | 6,252 | 2.46 |
| Italian | 0.432 | 182 | 0.510 | 2,690 | 3.13 |
| Spanish | 0.323 | 97 | 0.249 | 2,965 | 2.96 |
| Finnish | 0.029 | 68 | 0.039 | 2,889 | 2.95 |
| language | per-word fall | per word | per word |
|---|---|---|---|
| English | 3.158 | 3.358 | 0.200 |
| German | 2.623 | 2.735 | 0.112 |
| Italian | 1.415 | 1.426 | 0.011 |
| Spanish | 1.525 | 1.527 | 0.002 |
| Finnish | 0.083 | 0.085 | 0.002 |
| language | (full) | (matched) | (matched) |
|---|---|---|---|
| English, uniform sample | 27,162 | 9,508 | 1.005 |
| English, frequency-matched | 27,162 | 9,084 | 0.879 |
| German | 6,252 | 6,252 | 0.510 |
| Italian | 2,690 | 2,690 | 0.510 |
| Spanish | 2,965 | 2,965 | 0.249 |
| Finnish | 2,889 | 2,889 | 0.039 |
| check (script) | outcome measure | languages | statistic |
|---|---|---|---|
| Published depth bands ( run_caravolas.py ) | Caravolas (2022, p. 330) indicative bands | 8 that she places | 19 of 21 cross-band pairs ordered correctly, p = 0.014 by 20,000-permutation label shuffle; rho = 0.80, p = 0.010; with Danish’s stød marker deleted (§3.1), 20 of 21 and rho = 0.87; the same 19 of 21 and rho = 0.80 with eSpeak-ng English (§2) |
| Our measure ( ) against onset entropy ( run_seymour.py ) | Seymour et al. (2003) Grade 1 familiar content-word reading accuracy (their Table 5, the P1 rows, end of the first school year) | 6 — the whole intersection of Borgwaldt’s seven, Seymour’s groups and ours | onset entropy rho = 0.66 (n = 6; exact one-sided p = 0.088, two-sided 0.175), all-position rho = 0.31 (n = 6; one-sided p = 0.282, two-sided 0.564, the same with eSpeak-ng English, §2), exact over all 720 orderings |
| Position, holding corpus, aligner and units fixed ( run_position.py ) | The same Seymour et al. (2003) outcome | 11 | over non-initial alignment units rho = 0.75 (exact two-sided p = 0.010), over all units (the all-position measure) rho = 0.66 (n = 11; two-sided p = 0.031), over each word’s first alignment unit only rho = 0.64 (two-sided p = 0.040), exact over all 39,916,800 orderings; with eSpeak-ng English (§2) 0.75 (p = 0.010), 0.66 (p = 0.031) and 0.61 (p = 0.052) |
| Second outcome set ( run_ziegler.py ) | Ziegler et al. (2010) published phonological-awareness slope weights (their Table 2, p. 555) | 5, of which only Finnish is among the five this paper reports | onset entropy agrees better on three of their four outcomes and ties on decoding speed (reading speed, reading accuracy, decoding speed, decoding accuracy; a positive rho is agreement, higher entropy with a larger phonological-awareness weight): rho +0.80, +0.90, +0.67 and +0.97 against our +0.70, +0.50, +0.67 and +0.67 |
G2P Conversion
Grapheme-to-phoneme (G2P) conversion maps written text to pronunciations for speech technology and phonetic analysis.
See all 12 papers