4. Verb endings: -est, -st, -eth, -th

EModE verbs agree with their subject in more cells than modern verbs do. Modern English marks only the third person singular present (-s); EModE also marks the second person singular in the present and the past, and it has two competing third-person endings, -s and -eth. In the biblical register the old endings are kept everywhere: JPS 1917 never writes has or does (0 and 0 in the Torah), always hath and doth (2026-10-17).

4.1 The present indicative

History in one paragraph

-eth (Old English -eþ) was the southern third-person ending; -s came from the north. Hejná & Walkden (2022, §4.3.2) summarise: at 1500 -th is the most common form of the third person singular; it declines sharply during the period, and by 1700 it is rare, surviving longest in hath and doth; Shakespeare normally uses -s but keeps hath and doth (after Nevalainen & Raumolin-Brunberg 2016). The King James Bible (1611) is conservative and uses -eth throughout, and every later biblical revision copies it. So -eth in a 19th- or 20th-century text is a style marker, not a time marker.

Plural in -s and -th

Abbott §333: «Third person plural in -S. This form is extremely common in the Folio. It is generally altered by modern editors, so that its commonness has not been duly recognized.» His examples: «The venom clamours of a jealous woman / Poisons more deadly than a mad dog's tooth» (Comedy of Errors V.1.69). §334 gives plural -th: «Where men enforced doth speak anything» (Merchant of Venice III.2.33). §335: a singular verb before a plural subject, especially There is + plural, is normal.

For a parser this means: in Shakespeare, a verb in -s or -th does not prove a singular subject. Agreement cannot be used to pick the subject in EModE drama. In the biblical register it can: JPS 1917 agrees by the modern rule plus the 2nd person.

Second person -ts

Abbott §340: in verbs ending in -t, the 2nd person -test is often written -ts: «Thou torments» (Richard II IV.1, Folio), «Thou requests». A trace of the northern -s in the 2nd person.

4.2 The past

The 2nd person past adds -est or -st to the past form: camest, knewest, stoodest, broughtest, drankest, sentest, saidst, defiledst, stretchedst, madest, gavest, tookest. Wast is the indicative, wert the subjunctive (and later a free variant).

JPS 1917 examples:

• «Who told thee that thou wast naked?» (jps:GEN:71, GEN 3.11)
• «why didst thou not tell me that she was thy wife?» (jps:GEN:331, GEN 12.18)
• «and didst not suffer me to kiss my sons and my daughters?» (jps:GEN:965, GEN 31.28)

4.3 Spelling rules of the endings

(Counts «a+b»: tokens of the form with lemma = form in the db + tokens lemmatized correctly in the db, 2026-10-17; they add up to the frequency.)

doeth and doth, doest and dost. The full forms are the lexical verb do («he that doeth this»), the contracted forms the auxiliary («God doth know», jps:GEN:65, GEN 3.5). The distinction is a 19th-century editorial habit more than a 1611 rule, but JPS keeps it: doeth 15, doest 8 are always VERB in the db; doth 29, dost 7 are AUX 27 and 6 times.

False endings. Many ordinary words end in -eth, -est, -st or -th and must not be stripped:

morph::archaic (en/src/morph.rs) handles this with two guards: it acts only on a verb tag (VB, VBD, VBG, VBN, VBP, VBZ, MD), and it leaves alone any form the built-in dictionary already knows. The second guard is right for digest, interest, harvest and wrong for a set of AGID entries discussed in 4.6.

4.4 Negative and interrogative forms

The endings stay on the lexical verb when there is no do (chapter 7):

• «knowest thou not yet that Egypt is destroyed?» (jps:EXO:277, EXO 10.7) — inversion, the ending on the verb, not after the subject.
• «I know not» (jps:GEN:98, GEN 4.9).
• «Whither goest thou?» (tales and JPS: goest 27 tokens).

4.5 How Mova lemmatizes the endings

en::morph::archaic(lower, tag) (added 2026-10-17, commit 3e639a08e) runs first in Morph::lemmatize:

1. A list of 19 forms that are never anything but verbs (shalt, hath, hast, hadst, doth, dost, didst, wast, wert, canst, couldst, shouldst, wouldst, mayest, mayst, mightest, saith, sayest, spake) gets its lemma whatever the tag: the small model tags shalt as a noun.
2. With a verb tag only: 25 irregular forms (the list above plus wilt, art, brake, gavest, camest, wentest).
3. With a verb tag, for a form the dictionary does not know: strip -eth or -est, try the stem, the stem + e, the stem with a doubled final consonant undoubled, and -ieth → -y; return the first candidate that is a known lemma.

The unit test archaic_verbs checks knoweth, maketh, crieth, saith, and as negative controls art as a noun, greatest, rest, makes, digest, interest.

Measured against the db (2026-10-17)

I ran archaic and Morph::lemmatize (a small throw-away Rust binary linked to the en crate of this branch) over every VERB or AUX token of the db whose form ends in -eth, -th, -est, -st or -t and that is not an ordinary modern form, with the db's XPOS as the tag. 1 484 tokens:

The 400 unlemmatized tokens sit in 307 sentences with an -eth verb plus 21 sentences with a hath, hast or saith VERB (not AUX) whose lemma is the form (db, 2026-10-17). They did not pass through archaic with a verb tag; chapter 12 traces how (the small model tags them NOUN/ADJ, the graph parser makes them VERB and keeps the noun's lemma).

The lemmatizer's own errors (types, with tokens):

All three causes have deterministic fixes (chapter 13, items L1–L3):

1. take a candidate only if it is a verb lemma (has a verb paradigm), and among verb candidates prefer the most frequent in UD (put over putt, come over com, see over se);
2. ignore AGID-only JJS analyses when the tag is a verb tag, or drop AGID superlatives of non-adjectives (go, do, see, lie, should) from the dictionary;
3. for -est/-st/-edst after a past form (stood, brought, said, knew), look the stem up as a VBD form and return that form's lemma.

4.6 What the small model does with the endings

The tagger is the bigger problem. SLM now, aligned with the db (2026-10-17):

The model has never seen -eth, so it guesses from the shape: a long unknown word ending in -th is a noun (earth, death, breath). It has seen -est thousands of times — as the superlative. With the AGID entries above, the dictionary confirms the superlative reading, and the lemmatizer then strips -est by the superlative rule: camest → cam, comest → com, liest → ly, lest → l (63 of 66 lest, chapter 9).

Sentence counts, SLM now, out of 6 013 Torah sentences aligned with the db: 454 sentences have a mis-tagged -eth verb, 126 a mis-tagged 2nd person -est verb.

The graph parser and the teacher repair most of the UPOS (the db has the right UPOS), but the lemma survives (4.5) and so do wrong features (chapter 12).

4.7 Features

UD has no feature for «archaic ending», but Style=Arch is allowed on VERB and AUX by the registry. In the db:

Person and number of the 2nd person verbs are right in the db: VBP|Mood=Ind|Number=Sing|Person=2|Tense=Pres (110), VBD|…|Person=2|Tense=Past (19), with 3 stray Number=Plur|Person=3 copied from the tag's most frequent features. A deterministic rule can set them all: a verb whose form ends in a 2nd person ending and whose subject is thou (or no subject) gets Number=Sing|Person=2|Style=Arch; a verb in -eth/-th gets Number=Sing|Person=3|Style=Arch.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/en-x-emode/04-verb-endings/