bifonia

Pronunciation disambiguation for European-Portuguese heterophonic homographs. These are words spelled the same way whose pronunciation depends on meaning, not just part of speech. bifonia has zero runtime dependencies and uses only the Python standard library.

sede means thirst (ˈsedɨ, closed e) or headquarters (ˈsɛdɨ, open e). forma means mould (ˈfoɾmɐ) or shape (ˈfɔɾmɐ). molho means sauce (ˈmoʎu) or bundle (ˈmɔʎu). A text-to-speech front end that guesses wrong says the wrong word out loud. bifonia picks the right reading, and therefore the right IPA, from context.

from bifonia import tokenize, guess_sense, disambiguate

words = tokenize("Tinha tanta sede que bebi a garrafa toda.")
i = words.index("sede")
guess_sense(words, i)    # 'thirst'
disambiguate(words, i)   # 'ˈsedɨ'   (closed e)

words = tokenize("A sede da empresa fica em Lisboa.")
i = words.index("sede")
disambiguate(words, i)   # 'ˈsɛdɨ'   (open e, same spelling, different word)

Why part-of-speech tagging is not enough

The obvious approach is to tag the part of speech and pick the pronunciation from it. This does not work when two readings share a POS. sede thirst and seat are both nouns. forma mould and shape are both nouns.

A system can look like it solves this problem without solving it. Encyclopedic text (Wikipedia) is noun-skewed: one reading dominates and the minority readings barely occur, so a POS tagger that predicts the majority sense scores well on this text. The honest test is a sense-balanced set where the minority readings actually appear.

bifonia ships that test: bifonia-pt-homographs-gold, 2,346 real sentences (OpenSubtitles, Wikipedia, web pages), about 40% of them on POS-ambiguous readings, labelled independently of the engine.

Accuracy on the balanced real benchmark

A strong neural POS tagger drops to 74% on the POS-ambiguous half of the benchmark, since it has no signal to separate two senses that share a part of speech. A zero-dependency, meaning-keyed rule engine holds 97.6% on the same half. On the POS-separable half a tagger does well (about 93%); on the half that needs meaning, only meaning works.

A second finding stands on its own: curated word lists alone reach 87.3% on the hard cases, before any POS reasoning or learning. Meaning lives in collocations.

Two engines, one ensemble

The rule engine is self-contained and needs no training data, so it fits a low-resource language well. The learned models train from the labelled corpus and serve as a full-roster baseline and a corpus QC engine. guess_sense runs a per-word ensemble: each word is served by whichever engine wins on held-out behavioural data, so the combined system never scores worse than the rules alone.

A consistent result across this project: models trained on the corpus do not beat the rules out-of-distribution. The corpus is rule-labelled, so a model can at best mimic the rules in-distribution and generalizes worse on real text. This holds for Naive-Bayes, the averaged perceptron, and the classic Yarowsky decision list alike. The rules remain the strongest predictor, and an external POS tag can fuse on top:

# hybrid: trust an external tagger only where POS uniquely picks a sense, else rules
guess_sense(words, i, postag="NOUN")

On the encyclopedic wild set this hybrid ensemble reaches 95.3% (rules 90.7%, spaCy 91.6%). The tagger helps on POS-separable words, and the rules carry the rest.

Install

pip install bifonia          # zero dependencies, pure standard library

API

from bifonia import (tokenize, is_ambiguous, guess_sense, guess_pos,
                     disambiguate, add_extra_diacritics)

sentence = "Resolveu o problema desta forma simples."
words = tokenize(sentence)
i = words.index("forma")

guess_sense(words, i)                    # 'shape'
disambiguate(words, i)                   # 'ˈfɔɾmɐ'
disambiguate(words, i, sense="mould")    # 'ˈfoɾmɐ'  (override)
add_extra_diacritics(sentence)           # '...desta fórma simples.'  (acute = open vowel)

add_extra_diacritics rewrites each homograph with a disambiguating diacritic (acute for the open vowel, circumflex for the closed vowel) that a downstream grapheme-to-phoneme stage can read directly.

Datasets

All datasets are on the Hugging Face Hub, with schema {word, sense, pos, ipa, sentence}:

• bifonia-pt-homographs-gold: 2,346 real, sense-balanced, independently labelled sentences (43 words). This is the honest evaluation set.
• bifonia-pt-homographs: the labelled training and synthetic corpus with stratified splits.
• bifonia-pt-homographs-wild: the encyclopedic (noun-skewed) wild set.

Architecture

The design is layered and dependency-free, so a cue behaves the same way in the rules and in the model features:

bifonia/text.py      strip / accent-fold / proximity-weighted cue scoring   (shared, stdlib)
bifonia/cues.py      SENSE_CUES + FEATURE_CUES: the cue registry             (single source of truth)
bifonia/scoring.py   context rules + one generic cue resolver
bifonia/features.py  language-agnostic features, incl. CUE:<sense> from the registry
bifonia/locale/<lang>/*.voc   editable context + meaning wordlists
scripts/baselines.py         a Baseline protocol + zero-dep baselines for benchmarking

The meaning cues that disambiguate same-spelling readings live in one declarative registry (bifonia/cues.py), read by both the rule resolver and the model feature extractor. Adding a homograph is data-only: drop the .voc wordlist(s) and add one registry line, with no new code. Porting to a related language means supplying .voc files and, optionally, a corpus, since the algorithm carries no hardcoded Portuguese.

Coverage

131 homographs, 265 sense readings. The disambiguation axis is vowel quality (open ɔ/ɛ vs closed o/e). The hard core is the same-POS pairs a tagger cannot touch: sede, corte, forma, molho, bola, cor, lobo, polo, tola. Per-word IPA, senses, and diacritized forms are in docs/word_references.md.

Brazilian Portuguese

bifonia's pronunciations are European Portuguese, taken from Infopédia. A separate Brazilian table is not needed. For 122 of the 131 words, Brazilian Portuguese keeps the same contrast, with the same sense on the same stressed vowel. For 94 of these words, every sense was checked against the Brazilian transcription in Wiktionary. The others follow the general rule: nouns have a closed vowel and the homographic verb forms have an open one, as in o gosto [o] and eu gosto [ɔ]. One of the 122 words varies: Brazilian speakers say the noun esmero with [e] or [ɛ]. bifonia marks it esmêro, which is an accepted Brazilian form.

add_extra_diacritics marks the stressed-vowel quality with é/ê/ó/ô. The pt-BR lect of the phonemizer builds the other segments, such as final [i], palatalised t and d, coda s and r, and vocalised l. A mark can also change how the phonemizer reads the rest of the word. In tugaphone pt-BR-x-rj, abono gives [aˈbõnʊ] and abôno gives [aˈbonu].

Nine words differ in Brazil:

• para has [a] in both senses. bifonia marks the verb pára and leaves the preposition unmarked. This is also correct for Brazil.
• pelo has [e] as both contraction and noun. bifonia marks the noun pêlo and the verb pélo and leaves the contraction unmarked. This is also correct for Brazil.
• soma, abono, aceno, domo, empeno and desempeno have a closed nasal stressed vowel in both senses. Before m or n, Brazilian Portuguese for the most part has no open mid vowel.
• tola is closed in Brazil in the colloquial sense 'judgement, head' (Michaelis). bifonia files that sense and the timber under one head sense, and no Brazilian source covers the timber. So pt-BR keeps the European marks for tola.

Pass lang="pt-BR", or a sub-lect such as pt-BR-x-rj, to add_extra_diacritics to apply these differences. abono, aceno, domo, empeno and desempeno then stay unmarked. soma is always sôma, because an unmarked soma gets an open vowel from the tugalex pt-BR lexicons. Every other lect tag keeps the European marks, and None means the default. The rows are in bifonia/data/lect_overrides.csv, each with its source. For empeno and desempeno the source is the nasal rule and related verb forms, not a dictionary entry.

add_extra_diacritics("Ele soma os números.")                # 'Ele sóma os números.'
add_extra_diacritics("Ele soma os números.", lang="pt-BR")  # 'Ele sôma os números.'

Do not use the IPA column of heterophonic_homographs.csv as a transcription for Brazilian, African or Timorese lects.

African and Asian lects

The African and Asian lects pt-AO, pt-MZ, pt-CV, pt-GW, pt-ST, pt-TL and pt-MO keep the European marks, and bifonia has no override rows for them. No source found describes the Brazilian closing of a stressed mid vowel before m or n for any of these lects. For Angola and East Timor the sources point the other way: Luanda Portuguese has open nasal vowels (Cambolo & Osório 2022, pp. 203-205), and Timorese Portuguese does not nasalise a vowel before a nasal consonant (Albuquerque 2010, p. 278) and ontem is recorded with an open [ɔ] (Albuquerque 2011, p. 235). This evidence leads to an inference, not a sourced rule: the pt-BR rows do not apply to any of the seven lects. For pt-MZ, pt-CV, pt-GW, pt-ST and pt-MO the inference rests only on the absence of a source. Undolo (2014, pp. 183-185) and Cambolo & Osório (2022, pp. 198-200, with the summary on p. 210) report that Angolan speakers tend to open a stressed /e/ or /o/ where European Portuguese has a closed vowel, as in the noun desgosto and in sede 'thirst'. Both describe a variable tendency, not a categorical rule, so bifonia keeps the European marks for pt-AO.

Related projects

• TigreGotico/tugaphone: European Portuguese G2P, a consumer of bifonia's disambiguated output.
• TigreGotico/silabificador: European Portuguese syllabification.

See also

• docs/benchmarks.md: full benchmarks and subset plots
• docs/methodology.md: algorithm and features
• docs/usage.md: full API reference
• scripts/baselines.py · scripts/benchmark_ood.py: reproduce the numbers
• examples/basic_usage.py: runnable demo

License

Apache-2.0. See LICENSE.