> 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/yeddapalemeqblog.md). # Yedda Palemeq Blog ### Overview Yedda Palemeq, a Paiwan speaker and linguist, maintained a [blog](https://yeddapalemeq.blogspot.com/) with short Southern Paiwan examples, English translations, prose word explanations, and recordings. The corpus contains 668 source records represented as 671 sentences after three source-defined alternative splits. Some examples also occur in independently sourced FormosanBank collections; they are retained here with their blog provenance. **Many sentences are segmented into morphemes, but nothing is glossed at the morpheme level.** The `` tier records the author's own segmentation and carries no translations — it never has. Word-level glosses are prose explanations rather than interlinear glosses. See *Corpus Notes* below before using either tier. *** ### Corpus Statistics | |

Paiwan
Southern

| | ------------------------- | ------------------------- | | Word count | 5,877 | | Total audio | 1.2h | | Transcribed | 1.2h | | Untranscribed | 0 | | Translated words | | | English | 5,877 | | Mandarin | 0 | | Japanese | 0 | | Dutch | 0 | | Morphologically segmented | 4,475 | | Glossed words | 0 | *** ### **Access Details** * The corpus XML and retained processing code are available [in FormosanBank](https://github.com/FormosanBank/FormosanBank/tree/main/Corpora/YeddaPalemeqBlog). * `CodeAndDocs/scripts/` rebuilds the published XML from a frozen snapshot of the scrape, with no network access. `CodeAndDocs/scrape/` holds the code that read the blog in the first place; the build does not run it, and a fresh scrape of a live blog will not reproduce the snapshot byte-for-byte. *** ### **Corpus Notes** **Coverage** * The corpus covers the author's numbered "Paiwan Every Day" series, not the whole blog. The live Blogger feed holds 846 posts; the other posts are essays, announcements, and other non-series material that was never in scope. * The corpus holds 668 source records drawn from 641 post URLs, and the 2026-08-23 live-source audit matched every one of them against the feed. Six numbers in the series are nevertheless absent — 66, 183, 197, 480, 519, and 543. They have been missing since the corpus was first added and the reason is not recorded, so treat this as very nearly, but not quite, a complete run of the series. * The audit restored 24 incomplete or omitted sentence translations and 41 exact word glosses. Phrase-level definitions were not assigned to individual words. **Glossing** * Examples are not necessarily glossed word by word; repeated words and common function words are often unglossed in the source. Glosses were matched within a single example only — a gloss is never reused from another example, since it may not be contextually correct. Where no gloss can be found, the `` element's `
` is just a copy of the word in the example and no `` is provided. * Sometimes the same word is spelled differently in the main text and in the glosses. We have assumed the main text is correct and the gloss is incorrect; in those cases the gloss is ignored. * **This corpus is segmented but not glossed at the morpheme level.** The author glosses *words*, in prose, and never glosses individual morphemes, so no `` element carries a `` — and none ever has, in any version of this corpus. A morpheme has a form and a position and nothing else. That is why "Glossed words" above reads 0 while "Morphologically segmented" reads 4,475: segmentation and glossing are separate things here and only the first is present. Supplying morpheme glosses would be original linguistic work rather than data recovery. * Sixteen sentences carry no `` tier at all — the blog prints them as bare phrase entries with no word breakdown, so there is nothing to segment. They have no words and no morphemes. **Morpheme (``) elements** * Where the author parsed words into segments, those segments are preserved in `` elements. Within such a sentence, a word we could not find a parse for is assumed to be monomorphemic and gets a single `` — which is almost certainly incorrect in some cases. * Where the author parsed *nothing* in a sentence, that sentence carries **no `` elements at all**. An earlier version of this corpus gave every word a single `` mirroring the word, which asserted a monomorphemic analysis the author never made. Those mirror morphemes have been removed, so 1,165 of the 5,643 `` elements now have no `` child. This is why "Morphologically segmented" above counts 4,475 rather than the full word count. * Where a word contains an infix, the author marks it inline in the word itself with angle brackets — `seljec-an`, `pa-cun-an` — and that notation is what the `` tier is derived from. She then explains the infix in prose in the word's gloss rather than in Leipzig `` notation, so the analysis is present but written differently from most glossed corpora. * In short: the presence of an `` tier is itself information — it tells you the author analysed that sentence. This follows the bank-wide rule that morphological analysis is scoped per sentence, not per file; see [The FormosanBank XML format](/formosanbank/the-bank-architecture/formosanbank-xml-format.md) for POL-023. **Orthography** * The `original` tier keeps the source spelling exactly as the blog writes it. The `standard` tier is derived from it with FormosanBank's common conventions applied: source segmentation hyphens are removed (they belong on the `` and `` tiers), and diacritics that mark prosody rather than spelling are stripped. In this corpus that affects exactly two words — the Mandarin kin terms quoted in one sentence, where `yípó` and `āyí` become `yipo` and `ayi` in the standard tier only. Paiwan spells with no accented letter, so nothing here is protected; in a language that does, such as Puyuma with its `ē`, the letter survives. **Notation** * Infixes are written with angle brackets inside the word, as the source writes them: `seljec-an`, `pa-cun-an`. These are XML-escaped in the file (`s<em>eljec-an`), so any XML parser returns them as text. They are source notation, not markup — stripping tags naively will corrupt these words. **Alternatives, audio, and duplicates** * Where the source offers alternative wordings of one sentence — `qali/drava`, `tjangtjang / siyak`, `abar (yasi)` — each alternative becomes its own sentence record. The recordings for those three are excluded from the corpus: the author pronounces both alternatives in turn, which is not a recording of either single sentence and would be an ungrammatical utterance of the combined one. The files remain available as declared extras of the public audio dataset. * Two duplicate sentence groups come from separate blog posts with distinct source URLs and audio, so both attestations are retained. *** ### Copyright CC BY-NC 4.0, by permission of the author. Yedda Palemeq gave FormosanBank permission to license this material. The blog itself carries no public licence notice, so the licence rests on the author's grant rather than on anything stated on the site. *** ### 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: * Palemeq, Y. (2021). Yedda Palemeq. Retrieved May 19, 2026, from