Baka-ALCAM-MultimodalDataset

Language

Baka (ISO 639-3: bkc) is an Ubanguian language (Niger-Congo > Atlantic-Congo > Volta-Congo > Adamawa-Ubangi > Ubangi > Sere-Ngbaka-Mba > Ngbaka-Mba > Ngbaka > Western > Baka-Gundi) spoken by forest-based, hunter-gatherer communities of Cameroon and Gabon. According to the Administrative Atlas of Cameroon's Languages (Breton and Bikia Fohtung 1991), Baka belongs to the Ubanguian speech area, with its major settlements located in the Boumba-et-Ngoko Division of the East Region of Cameroon; the language is also spoken in the Kadey and Haut-Nyong divisions of the East Region and in the Dja-et-Lobo division of the South Region. Ethnologue estimates approximately 40,000 speakers in Cameroon, of whom around 15,000 are monolingual, with a language vitality status of "vigorous" (EGIDS 6a); Glottolog similarly classifies the Cameroonian variety as showing sound cross-generational transmission, with most speakers also bilingual in the majority language(s) of their area. The Baka have historically faced social marginalization from neighboring Bantu populations — a dynamic that outside sources have sometimes framed in terms of the exonymic and now widely contested "Pygmy" label, which is used here only as an attribution from such sources and not as an identification of the community — and increasing regional population density and Baka sedentism have contributed to rising bilingualism and, in some areas, gradual shift toward French as the administrative and educational lingua franca.

Variants

The present release does not distinguish internal dialectal variation within Baka; the datasheet and audio recordings represent a single, undifferentiated variety as elicited from one native speaker. Unlike the multi-dialect structure adopted for some other languages in the same ALCAM series (e.g. Batanga, which documents the distinct Banoho and Bapuku varieties separately), no dialect-specific split has been documented for Baka in this dataset.

Writing System

The writing system used for the transcription of Baka in this dataset is the International Phonetic Alphabet (IPA), as reflected in the Word, LangEx and LangPars columns of the datasheet and in the sentence field of the audio-sentence mapping files. The phonological inventory below draws on the 373 datasheet entries and the 247 unique audio-aligned sentences (the mapping.tsv files across the dataset's recording batches).

1. Vowels

The data attests seven oral vowel qualities: a, e, ɛ, i, o, ɔ, u. No systematic nasalization marking (e.g. combining tilde) is attested in the transcribed data, in contrast to some other languages documented in the same ALCAM series.

2. Consonants

The consonant inventory attested in the datasheet and mapping files includes:

b, d, g, j, k, l, m, n, p, s, t, w, y — the core set of simple consonants, plus two implosives, ɓ (voiced bilabial implosive) and ɗ (voiced alveolar implosive), which occur only rarely across the datasheet and mapping files (12 and 5 tokens respectively).

Complex onsets and prenasalized consonants are also attested: mb, nd, ng, ny, kp, gb, and the less frequent mgb (prenasalized labial-velar). An apostrophe (') also occurs within several transcribed sentences, most plausibly marking a glottal stop or a clitic/word boundary; its precise grammatical function has not been formally documented.

One additional, non-standard character — ȹ (Latin small letter "qp" digraph, U+0239) — recurs consistently across a number of specific lexical items and their example sentences (e.g. ȹē, ȹɛ, ȹítì, ȹàȹālā). It is used independently of the ny digraph, which is separately and consistently attested elsewhere in the data. Its intended phonetic value in this transcription is not stated in the source materials and should be clarified with the dataset's creator before downstream use.

3. Tone system

Three level tones are marked in the datasheet and mapping files:

• High tone (H): á, é, ɛ́, í, ó, ɔ́, ú

• Mid tone (M): ā, ē, ɛ̄, ī, ō, ɔ̄, ū

• Low tone (L): à, è, ɛ̀, ì, ò, ɔ̀, ù

Unmarked vowels represent tonally neutral or contextually determined syllables. No contour-tone diacritics (rising, falling or extra-low) are attested in the transcribed material, unlike in some other datasets of the same ALCAM series.

Source

The dataset was collected through a questionnaire designed to gather basic information about the Baka lexicon and grammar. This was done as part of the Atlas Linguistique du Cameroun (ALCAM) project.

Domain

The dataset represents a linguistic questionnaire designed to elicit the basic lexicon and grammatical information of the Baka language.

Size

The datasheet comprises 373 elicited lexical/example-sentence entries. The audio subfolder contains 315 audio files distributed across four recording batches (68, 81, 96 and 70 clips respectively; 4m 46s, 4m 05s, 6m 26s and 4m 53s of speech, measured directly from the audio files). The first batch (68 clips) is fully redundant with the fourth batch: 68 of the fourth batch's 70 files share the same filenames and identical transcribed sentences as the first batch, indicating a re-recording or continuation session rather than independent content. Discounting this redundancy, the dataset contains 247 unique sentence-audio pairs, corresponding to 15 minutes 28 seconds of speech; including the duplicated first batch, the four raw recording batches total 20 minutes 12 seconds of audio across 315 files.

Structure

The dataset comprises: 1) a datasheet (Baka-ALCAM-MultimodalDataset.tsv) with lexical entries, French glosses, and example sentences with word-for-word parsing; 2) voice clips read by a native speaker of Baka, organized into four recording batches under the audio subfolder; 3) sentence-to-audio mapping files (mapping.tsv, one per recording batch).

Description of columns

Datasheet (Baka-ALCAM-MultimodalDataset.tsv):

• OrigID: original number of lexical entry on paper questionnaire

• EditID: modification of OrigID

• FrenchRef: reference entry (originally provided in French)

• FrenchComm: original comments about reference entry (FrenchRef)

• French: lexical entry in French (overlaps with FrenchRef)

• Note: note of researcher on the lexical entry

• POS: part of speech

• Class: noun class (where applicable); not populated in this release, as Baka, an Ubanguian language, does not have a Bantu-style noun-class system

• Morf: morphological attribute (e.g. plural, singular)

• Var: dialect variant tag; not populated in this release, since the dataset represents a single, undifferentiated variety (see Variants above)

• Word: lexical entry in Baka (IPA)

• CrossRef: cross-referencing of lexical entry number

• FrenchEx: example sentence in French

• LangEx: example sentence in Baka (IPA)

• LangExEdit: manual editing of LangEx

• FrenchExEdit: edited French equivalent of FrenchEx

• LangPars: word-for-word parsing in Baka

• LangParsEdit: editing of LangPars

• FrenchPars: French equivalent of LangParsEdit

• FrenchParsEdit: editing of FrenchPars

_ marks fields left empty in the source questionnaire. Of the 373 entries, 373 have a French gloss, 325 have a Baka Word, 259 have a French example sentence, 256 have a Baka example sentence, and 244 have word-for-word parsing. The audio subfolder separately provides recordings for 247 unique sentence-level items overall (see Size, below); a spot-check confirms several of these correspond exactly to LangEx entries in the datasheet (see Sample, below), but an exhaustive row-by-row correspondence between every audio clip and a specific datasheet entry has not been established.

Mapping files (mapping.tsv):

• audio_filename: name of the corresponding audio file

• key: audio filename without its extension

• sentence: the elicited word and its example sentence, separated by a semicolon (word ; example)

• attempts: number of recording attempts (takes) for that item; 295 of the 315 recorded items were captured in a single attempt, with the remainder requiring two to five attempts

Sample

The excerpt below shows 5 of the 373 datasheet entries (OrigID x48-x53) for which the Baka example sentence (LangEx) has an exact, verified match in one of the audio-sentence mapping files, illustrating the parallel Baka/French structure together with the aligned audio.