10. Vocabulary: archaic words and false friends

Most of the EModE lexicon is still in use, which is why the period reads easily. The danger is in two kinds of words:

• archaic words that have left the language (peradventure, wherewith, kine, murrain) — a parser meets them as unknown words and guesses their part of speech from their shape;
• false friends that are still in use with another meaning (even = evening, corn = grain, suffer = allow, quick = alive, meat = food, let = hinder). These are worse: the parser is sure of itself, the tree is fine, and the world built from it is wrong.

This chapter is a working lexicon for the biblical register and the tales: form, part of speech, EModE meaning, an example with its id, and what Mova does now. Counts are from JPS 1917 (2026-10-17) unless marked.

10.1 Pro-adverbs of place, direction and reference

EModE (and Old English) build adverbs from here, there, where + a preposition and use -ce/-ther forms for direction. They are closed-class and regular.

UD. All are ADV with advmod, except when they head a relative clause (wherein, whereof, whereon, wherewith, whereby introduce relative clauses: in UD they are ADV, advmod or obl inside the relative clause, and the clause is acl:relcl of the antecedent). PronType=Rel or PronType=Int may be added.

Mova now. Who can do what to whom — level-1 selectional preferences (2026-10-02) defines a selection archaic_adverbs used by the reranker (world/src/rerank.rs, repair ArchaicAdverb: «noun → ADV advmod») and by rule induction (world/src/induce.rs, feature F_ARCHAIC): whence whither thence hither thither hence yonder wherefore thereupon whereupon forthwith. The list misses the most frequent ones in the Torah: thereof (361), therein (64), wherewith (47), wherein (36), thereon (26), nigh (29), peradventure (18), whereon (13), howbeit (12), whereof (10), therewith (10), whereby (9), thereby (6), henceforth (4), hitherto, thenceforth, betimes, yesternight, afar. Of 6 013 aligned Torah sentences, 473 have a UPOS error on one of these adverbs (SLM now). Chapter 13, T2.

10.2 Discourse words and particles

10.3 False friends

The meaning in the Torah differs from the modern one. The tree is right; the frame is wrong if the modern meaning is used.

Sources for the KJV-only false friends: W. A. Wright, The Bible Word-Book (2nd ed. 1884, public domain), which lists such words with their 1611 senses (listed under «wanted» in Sources: not yet in our library); Hejná & Walkden 2022 ch. 7 on semantic change; the examples themselves are verified in the texts cited.

10.4 Archaic nouns, adjectives and verbs (no modern homograph)

10.5 Measures, ritual and law words

These are not archaic in the sense of the period — they are technical — but they are the words a modern parser does not know as nouns: shekel 33, cubit 20, ephah 18, hin 16, omer 3, tithe 13, wave-offering 15, heave-offering, sin-offering 99, burnt-offering 148, meal-offering 100, peace-offerings 54, first-fruits 14, atonement 79, unclean 142, abomination 18, statute 30, ordinance 19, testimony 35. The SLM gets hin (a liquid measure, 16) right as a NOUN, which shows that a word unknown to modern English is not the problem when its shape is ordinary; the Judaism dictionary of the world layer (world/src/dictionary/judaism.rs) is the place to give these words their records.

10.6 What to put in the lexicon

1. Closed class, fixed analysis (no context needed): the 30-odd pro-adverbs of 10.1, the discourse words of 10.2, the pronouns of chapter 3 and the auxiliaries of chapter 5. A table in front of the tagger.
2. Open class, lemma only: the old plurals and strong verbs (kine → cow, brethren → brother, sod → seethe). A table in the lemmatizer, applied with the right tag.
3. False friends: nothing in the parser. They need a register-dependent sense in the world dictionary: when the text is in the biblical register (the Style=Arch density of the document, or its source), corn means grain, even after at means evening, suffer + object + infinitive means allow, know + a person as object + conception in the next clause means a union. world/src/domains.rs already has a «formal» domain that includes senses labelled archaic or literary; the false friends above are a seed list for it.

10.7 The dictionary layer (world dictionary emode)

The lists above are the seed of a dictionary layer for the pupil: world/src/dictionary/emode.rs finds every archaic word, form and sense of the archaic corpora (the JPS Torah, the 0.3 tales, the English Sefer Yetzirah) and writes world/data/dictionary-schools/emode.tsv (pages in emode/ by world dictionary schools). Each record: the form, its kind (archaic form / word / sense), the modern equivalent, part of speech, the meaning a pupil needs, how it differs from today, up to three corpus sentences with their source, and the tests that took it in, with their numbers. The base dictionary is not touched.

world dictionary emode requests <work> --titles <tales books.tsv> torah.tsv pool.tsv sy.tsv
annot batch-run <work>                       # Haiku 5.5 through the Batches API
world dictionary emode post <work>           # → emode-senses.tsv
world dictionary schools <out>               # → emode.tsv and the pages

The tests (2026-10-18; 140 555 sentences, 3 055 469 tokens, 29 548 forms; EWT 173 060 tokens):

The rate test only proposes: Wiktionary's first sense is often just the oldest (horrid = bristling), and an archaic sense somewhere says nothing of our sentences; a first version over all the corpora took 52 words, 47 of them tales' topics (princess, palace, tree). The teacher therefore also answers «old here: yes/no», and a rate word stays only on yes: 97 of 118 stayed. The ending forms get no teacher: their record is built from the stem's record. The teacher wrote the senses of 319 words (8 requests, one retried with more tokens: 78 856 input and 87 819 output tokens). Result: 489 records (251 archaic forms, 141 archaic words, 97 archaic senses). Two runs give the same files, byte for byte.

Limits: the teacher is lenient on the rate words (it kept basket, voice, tent for narrow old uses); a manual false friend stays even when the corpora use only today's sense (let, prevent: the record says so, though for prevent the teacher wrongly claims the old sense); the OCR'd Westcott 1893 brings some library boilerplate into the examples. Wiktionary's labels (CC BY-SA) are used only as a test; nothing of their text is carried.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/en-x-emode/10-vocabulary/