Linked Open Data

The tag layer, expressed in two standards that are not ours — so that other people's tools can read it without knowing anything about this project.

The short version. This project normalises a verbatim sentence in one discipline to a shared tag. A gazetteer project like MEHDIE or Kima normalises a place name in one language to a shared identifier. Same operation, different axis — so the same two standards fit both:

SKOS turns each tag into an addressable concept with a URI, labels in English and Hebrew, and a link to Wikidata.
W3C Web Annotation turns a grounded tag into an annotation whose target carries a TextQuoteSelector. That selector — exact, prefix, suffix — is this project's verbatim-grounding rule. The house rule turned out to be a W3C rule. This export publishes the five content_tags facets, which hold 34,121 of the corpus’s 118,615 live quote fields (29%). The other layers’ quotes — concepts, limitations, concept edges, mechanisms, senses and the rest — are not annotations here yet, though their terms are in the vocabulary.

Files

vocab.html — browse the vocabulary vocab.jsonld 11.6 MB · 525 KB gzipped vocab.ttl 9.0 MB · 505 KB gzipped annotations.jsonld 81.7 MB · 5.2 MB gzipped corpus.jsonld 5.2 MB · 616 KB gzipped

Scope — read this before comparing counts. This export is generated from the project's working corpus (1698 records; built from DATA/corpus.json, corpus md5 050fd81816850c73), which is larger than the frozen release corpus published at DATA/corpus.json (126 records) and uses a fuller schema. The two will not agree on totals, and are not meant to. Where a record has a published page the annotations target that page; where it does not, they target the paper's DOI. Regenerate with python TOOLS/export_lod.py --validate.

What is in here

Concept scheme	Concepts	Linked
Disciplines	211	39
Methods	34	13
Domain concepts	22829	77
Statistical terms	1157	25
Places	70	64

34285 annotations over 34383 verbatim spans. Each annotation carries up to two anchors, and they are not the same thing.

Of the 15,165 annotations on papers that are not cleared, 12,538 carry a position (a start and an end) and 2,627 carry none: a position is published only where the quote was located. 2,013 annotations on papers that are not cleared carry no wording of ours, because our wording repeated part of the paper’s text; each says so (mt:bodyWithheld). 1,687 of them carry a position, 326 carry none. 94 annotations carry mt:unvouched: their quote could not be confirmed mechanically as an exact substring of our copy of the paper, and the value says why.

The source anchor is the position of the quote in the paper's own staged text — the text every quote in this corpus is verified against — and it is the one to resolve against the article. 25482 of the spans have one, 25410 of those from offsets this project had already verified. 8,901 have none: for 8,818 the quote, exactly as this corpus stores it, is in that text: this export looked for a re-spelled form of the quote (curly quotation marks, dashes and ligatures made plain) and did not find that form, so the position the corpus holds for it is not published here yet; for 83 the quote as stored is not in that text as one string.

The display anchor is the position of the same quote in this project's own page for the paper, so a reader can be taken to the spot. It is not provenance: the pages interleave the paper's sentences with our tag layer, so a window of text around a quote on our page may be our words. Where it was, 15985 of those windows were dropped and 3160 kept.

Of the 19,187 spans on papers whose text we are cleared to publish, 19,145 were found on the paper’s page (99.8%) and 42 were not. The other 15,196 spans are on papers whose text is not cleared. Their pages carry our tags and short quotations, not the article, so a quote is found there only where the page itself prints it: 2,881 were found, 12,315 were not. For those papers an annotation never carries the sentence. It carries a position where one was measured and none where it was not; the paragraph on wording above gives the counts.

Two things this export deliberately does not contain. The reliability layer is private and is never published. And paper-level facets are statements about the paper rather than annotations: the discipline and the concepts are dcterms:subject, the method is mt:method, and the other paper-level facets are not exported. That is not because they have no quote behind them. 8,782 paper-level evidence quotes of 40 or more characters are recorded on the exported records, and every one of them is an exact substring of the paper’s own text. The reason is narrower: the annotation model here is scoped to the five content_tags facets, and the evidence behind a paper-level facet is not modelled as a target yet — work not done, not a property of the data. Whether it should be published is a question about the scope of this dataset, not about whether the quotes exist.

One annotation, in full

{
 "id": "https://shir-openu.github.io/meta-tagging-showcase/lod/annotations.jsonld#mendel1866-plant-hybridization--findings-0",
 "type": "Annotation",
 "motivation": "describing",
 "mt:unvouched": "declared-ellipsis",
 "mt:quotedInPieces": 2,
 "created": "2026-07-15",
 "creator": {
  "id": "https://orcid.org/0009-0000-2597-2824",
  "type": "Person",
  "name": "Shir Sivroni",
  "nickname": "Shir Sivroni"
 },
 "generator": {
  "id": "https://shir-openu.github.io/meta-tagging-showcase/",
  "type": "Software",
  "name": "Meta-Tagging Project tag layer"
 },
 "body": [
  {
   "type": "TextualBody",
   "purpose": "describing",
   "value": "Crossing pea (Pisum) plants with constant differing characters, hybrids show one character (dominant) while the other becomes latent (recessive); in the next generation dominant and recessive forms appear in an average ratio of 2.98:1 ≈ 3:1.",
   "language": "en"
  }
 ],
 "target": [
  {
   "type": "SpecificResource",
   "source": "https://shir-openu.github.io/meta-tagging-showcase/#staged/mendel1866-plant-hybridization",
   "mt:role": "source-anchor",
   "mt:anchorText": "the staged body of this paper, the text every quote in this corpus is verified against",
   "mt:sourceSha1": "d18eeb1aee9c697a1127a6cb4563657f9305ad98",
   "mt:extractedBy": "the staged body of the paper, as ingested; verify_anchors.body(pid)",
   "selector": {
    "type": "TextQuoteSelector",
    "exact": "are termed the dominant, and those which become latent in the process recessive",
    "prefix": "e the characters of the hybrid, ",
    "suffix": ". The expression recessive has b",
    "start": 16937,
    "end": 17016
   }
  },
  {
   "type": "SpecificResource",
   "source": "https://shir-openu.github.io/meta-tagging-showcase/#page-text/mendel1866-plant-hybridization",
   "mt:role": "display-anchor",
   "mt:anchorText": "THE EXTRACTED TEXT of this project's page for the paper - the page with its scripts, its markup and our tag chips removed, which is the string these offsets were measured on and is not the bytes of the HTML file. The window around the quote may contain the tag layer, so it carries no context here.",
   "selector": {
    "type": "TextQuoteSelector",
    "exact": "are termed the dominant, and those which become latent in the process recessive",
    "start": 7611,
    "end": 7690
   },
   "mt:sourceSha1": "3c65024e79b4b236dd343aa31a14002566ed5b5a",
   "mt:extractedBy": "export_lod.page_text (strips script/style, removes the tag chips, drops every tag, then norm_text); export_lod.py sha1 1fb9e1da59106cc1348ebfc10e3fbdfae2faf6c7",
   "mt:renderedAt": "https://shir-openu.github.io/meta-tagging-showcase/papers/mendel1866-plant-hybridization.html"
  },
  {
   "type": "SpecificResource",
   "source": "https://shir-openu.github.io/meta-tagging-showcase/#staged/mendel1866-plant-hybridization",
   "mt:role": "source-anchor",
   "mt:anchorText": "the staged body of this paper, the text every quote in this corpus is verified against",
   "mt:sourceSha1": "d18eeb1aee9c697a1127a6cb4563657f9305ad98",
   "mt:extractedBy": "the staged body of the paper, as ingested; verify_anchors.body(pid)",
   "selector": {
    "type": "TextQuoteSelector",
    "exact": "an average ratio of 2.98:1, or 3:1",
    "prefix": "inant and recessive characters, ",
    "suffix": ". The dominant character can hav",
    "start": 24832,
    "end": 24866
   }
  },
  {
   "type": "SpecificResource",
   "source": "https://shir-openu.github.io/meta-tagging-showcase/#page-text/mendel1866-plant-hybridization",
   "mt:role": "display-anchor",
   "mt:anchorText": "THE EXTRACTED TEXT of this project's page for the paper - the page with its scripts, its markup and our tag chips removed, which is the string these offsets were measured on and is not the bytes of the HTML file. The window around the quote may contain the tag layer, so it carries no context here.",
   "selector": {
    "type": "TextQuoteSelector",
    "exact": "an average ratio of 2.98:1, or 3:1",
    "prefix": "inant and recessive characters, ",
    "suffix": ".\") and the independence stateme",
    "start": 862,
    "end": 896
   },
   "mt:sourceSha1": "3c65024e79b4b236dd343aa31a14002566ed5b5a",
   "mt:extractedBy": "export_lod.page_text (strips script/style, removes the tag chips, drops every tag, then norm_text); export_lod.py sha1 1fb9e1da59106cc1348ebfc10e3fbdfae2faf6c7",
   "mt:renderedAt": "https://shir-openu.github.io/meta-tagging-showcase/papers/mendel1866-plant-hybridization.html"
  }
 ]
}

Reconciliation was not automatic, and says so

A link to an outside authority is asserted on one of three grounds. For 64 of the 218 linked terms: a label or alias that matched, and a Wikidata type statement (P31 or P279) licensing it. For 63: a person’s decision, recorded with the term as ‘human adjudication’; no type statement licensed it. For 91: a matching label alone, in groups for which no type list was defined. Where none of these could be shown, the result is recorded as unmatched rather than guessed. Place names needed a further step — Cambridge is two of the most important academic cities in the world, and no string test settles which one a paper means. Those were decided by hand, one at a time, with the reason recorded next to the decision, and one string ('Cambridge') had to be decided per paper rather than once. This is the same wall MEHDIE hits and answers with a reconciliation review interface.

Author, rights and how to cite

Shir Sivroni — ORCID 0009-0000-2597-2824
Department of Mathematics and Computer Science, The Open University of Israel

Licence: CC BY 4.0. You may reuse and redistribute this, including commercially, provided you attribute it.

Archived and citable: https://doi.org/10.5281/zenodo.21432188 (Zenodo concept DOI — always resolves to the latest version)
Source: https://github.com/shir-openu/meta-tagging-showcase

Cite as:
Sivroni, S. A Verbatim-Grounded, Field-Neutral Tag Layer for Cross-Disciplinary Reading of Academic Literature. doi:10.5281/zenodo.21432188 (CC BY 4.0).

This attribution is carried inside the data as well as on this page: every annotation names the ORCID, and each file carries the licence, the DOI and the repository as triples — so a consumer who keeps only the graph still knows whose work it is and how to credit it.