> For the complete documentation index, see [llms.txt](https://ai4commsci.gitbook.io/formosanbank/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://ai4commsci.gitbook.io/formosanbank/the-bank-architecture/corpora/song-kanakanavu-grammar.md).

# Song (2018) Kanakanavu Grammar

This corpus is the Kanakanavu (`xnb`) material from Song Limei's 2018 *Kanakanafu yu yufa gailun* \[Introduction to Kanakanavu Grammar] (Council of Indigenous Peoples), scraped from the official Alilin e-reader and reconciled page-by-page against the source page images. It comprises 699 grammar sentences — 650 of them with word- and morpheme-level interlinear glossing (3,477 words, 5,048 morphemes) — together with the book's Appendix 2A dictionary as a separate 870-entry lexicon. Text is in the Ortho113 orthography with Mandarin (`zho`) translations and glosses.

***

## Corpus Statistics

|                           | Kanakanavu |
| ------------------------- | ---------- |
| Word count                | 5,256      |
| Total audio               | 0          |
| Transcribed               | 0          |
| Untranscribed             | 0          |
| Translated words          |            |
| English                   | 0          |
| Mandarin                  | 5,256      |
| Japanese                  | 0          |
| Dutch                     | 0          |
| Morphologically segmented | 3,999      |
| Glossed words             | 3,999      |

***

## **Corpus Processing**

The text was scraped from the official Alilin e-reader and reconciled page by page against the source page images. The book's own orthography discussion (reader pp. 31–44) describes the Ortho113 inventory — `ʉ` and `r`, no `l` — so no orthographic conversion is applied: the standard tier differs from the original tier only where the source's own apparatus does not belong in a standard surface.

Every departure from the printed text is recorded against a named record and a reader page; nothing is changed by a blanket rule. The full list is in the corpus [README](https://github.com/FormosanBank/FormosanBank/blob/main/Corpora/Song-Kanakanavu-Grammar/README.md#recorded-per-record-corrections), and the reasoning behind the standard-tier decisions is in [`standard_surface_review.md`](https://github.com/FormosanBank/FormosanBank/blob/main/Corpora/Song-Kanakanavu-Grammar/CodeAndDocs/docs/standard_surface_review.md).

***

## **Corpus Notes**

* **Stress accents** (`á é í ó ú`) mark stress in the book, not distinct letters. They are kept in the original tier and folded out of the standard tier; PHON is accent-free on both tiers.
* **The standard tier is nearly the original tier.** The book's orthography already *is* Ortho113, so nothing is transliterated: across the whole corpus the standard tier differs from the original only by the folded stress accents and by the one song example below. Do not read "standard" here as evidence of a spelling change.
* **Everything is in Mandarin; there is no English.** All 10,090 translations and glosses are `xml:lang="zho"`. Morpheme glosses use the book's own Chinese category labels — `主事焦點` (actor focus), `受事焦點` (patient focus), `完成貌` (perfective), `非實現` (irrealis), `重疊` (reduplication) — rather than Leipzig abbreviations. Latin script inside a gloss is a proper name.
* **Segmentation lives on the word and morpheme tiers, never on the sentence tier.** Sentence forms are running text on both tiers. Word and morpheme forms keep the source's `-`, `=` and `<…>` on both tiers.
* **Phonetic tiers use pipe notation for phonemic variants.** Ortho113 maps Kanakanavu `r` to `[r|ɾ]`, so a bracketed pipe in PHON means "either of these", not a literal bracket. PHON is otherwise marker-free and accent-free.
* **No audio.** The source is a printed book; no recordings exist for this corpus.
* **Interlinear conventions.** W forms keep the source's `-`, `=` and `<…>` notation. An infixed root keeps a gap hyphen and the infix is written `-X-` (`t<um>a-túturu` → `t-a`, `-um-`, `túturu`); each clitic morpheme keeps a leading `=` (`te=musu` → `te`, `=musu`).
* **Partial glossing is preserved, not invented.** On five page images the printed form and gloss lines do not align one-to-one. Those words are still segmented, but a morpheme the book does not gloss separately carries no gloss of its own; the word-level gloss is always present.
* **One repaired typesetting slip.** On reader p. 69 the aligned analysis line prints `takanaga` where the sentence line of the same example prints `takananga` (`takananga` occurs 22× in the corpus, `takanaga` twice — both inside that one analysis). It is repaired in the original tier as a recorded manual edit, so the standard tier and the phonetic tiers regenerate from the corrected form; the extraction ledgers still record exactly what the page prints.
* **Three examples come from the book's punctuation table (Appendix 1, p. 190), and its marks are not what they look like.**
  * *The song.* Row 10 of that table defines a hyphen used **in songs, to break lyrics to fit the musical score** — not morpheme boundaries. The one song example keeps those hyphens exactly as printed in the original tier. Its standard tier gives the running words, whose boundaries were recovered from an independent attestation of the same song: deleting the hyphens is not enough, since the last line would run together into a single word. This is the only place in the corpus where the standard tier is set from a source-checked decision rather than derived.
  * *The dash.* Row 8 defines `--`, written with two hyphens, as the Kanakanavu spelling of the **em dash** (破折號), whose Chinese equivalent is `―`. It is punctuation introducing an explanation or a list, not segmentation. The build renders it as a single `-` in the original tier, which the standard tier then inherits unchanged — so the two examples using it show a `-` that is a dash, not a morpheme boundary. They are the corpus's only two occurrences of a hyphen in a sentence-level standard form.
* **Dictionary variants are split, not concatenated.** Slash and semicolon alternatives and optional `(…)` material in Appendix 2A become separate single-form entries. Bound citation forms (a trailing hyphen, with no free-standing surface in the book) are excluded, and split variants that duplicate another record are dropped. Every exclusion is a recorded decision.
* **Repeated forms are deliberate, and are distinct attestations.** 63 groups recur: 28 are grammar examples the book reuses to illustrate different points, 34 are dictionary homographs with different translations, and one is a pair of stress variants that fold to the same standard form. Each `S` carries its own `source` attribute naming the page and label it came from. There is no exact same-locator duplicate, and nothing is deduplicated.
* **49 of the 699 sentences have no interlinear analysis.** 47 are the punctuation examples of Appendix 1 (pp. 187–191, a punctuation table rather than a glossed text) and 2 are body examples the book leaves unglossed. The book prints no word gloss for any of them, and none was invented.
* **14 source examples were excluded**: 6 noun-phrase fragments, 5 forms the book marks ungrammatical, and 3 examples printed without a Chinese translation. Each exclusion is recorded with its reason in the corpus's source ledger.
* **Everything above is checkable.** No validator reports a HARD finding. The remaining warnings are the documented source properties on this page — the unglossed sentences, the two em dashes, the four unglossed morphemes, the host+clitic word count, and the repeated forms — and the corpus README lists them one by one with counts.

***

## **Access Details**

* The repo containing this corpus in FormosanBank as well as the code to reconstruct the corpus can be found [here](https://github.com/FormosanBank/FormosanBank/tree/main/Corpora/Song-Kanakanavu-Grammar).

***

## **Copyright**

CC BY-NC 4.0 (Creative Commons Attribution–NonCommercial), by permission of the author (Li-May Sung).

***

## Citation

In accordance with our [Terms of Use](/formosanbank/additional-resources/terms-of-use.md), if you use this corpus or any product derived from this corpus in any publication, you must cite both FormosanBank and:

Song, Limei. (2018). *Kanakanafu yu yufa gailun* \[Introduction to Kanakanavu Grammar]. *Taiwan nandao yuyan congshu*, 16. Council of Indigenous Peoples.