📄 The two be 's of English

Summary

Payne examines how be is used in Contemporary English. He questions the textbook split between a "lexical" copula (She is a doctor) and an "auxiliary" be (She is waiting). Using standard morphosyntactic diagnostics, he argues that the copular use is a fully fledged core auxiliary. These diagnostics are cliticization and the NICE properties (Negation, Inversion, Code, Emphasis). On every test, copular be behaves like modals, have and progressive or passive be. True linking verbs like seem, become and resemble behave the opposite way and need do-support.

He then proposes that English does have a lexical be, but the dividing line is semantic rather than copular versus non-copular. That lexical be is an active, agentive verb, roughly "act (like)". It patterns with ordinary verbs, and for some speakers it is even inflected regularly.

The broader consequence is that an English clause does not need a lexical verb, only Inflection. In copular clauses the complement carries the predication. The author argues this view simplifies both grammatical theory and English teaching.

Key points

• Diagnostic battery: the tests are the NICE properties from Huddleston & Pullum (2002, pp. 92–112) plus contraction onto a preceding host.
  • Contraction works with locative fronting too: In the kitchen's a sink.
  • Lexical verbs fail contraction: *The teacher's a Jaguar is impossible for "has".
• Copula versus linking verbs:
  • Copular be passes all tests without do: Is she a doctor? is fine, *Does she be a doctor? is not.
  • Seem, become, resemble and turn require do, so they are lexical main verbs.
• Second claim (active be):
  • Agentive be is a separate lexical verb, paraphrasable as act/act like.
  • Some speakers give it regular forms, be's and beed.
  • It groups with lexical linking verbs, not with the auxiliary.
• Clause requirement: the paper replaces "every sentence needs a lexical verb" with "every sentence needs Inflection". Copular, passive and progressive constructions are unified as taking auxiliary be.
• Cross-linguistic and corpus notes:
  • Mandarin splits the functions of be across shì, zài and yŏu.
  • Spanish splits them across ser, estar and haber.
  • The British National Corpus is consulted for copula omission in whence constructions.

Interesting ideas

• I1. Categories are assigned by behaviour across a fixed set of syntactic tests (negation placement, subject inversion, tag/ellipsis, verum stress, contraction), not by meaning or tradition, which makes the classification checkable and reproducible.
• I2. In a copular clause the predicating element is the nominal, adjectival or locative complement, while be only hosts tense and agreement. This is a clean division of labour between content and inflection.
• I3. One surface form, be, splits into a stative auxiliary and an active agentive verb ("act like"), and the split is motivated by stativity versus activity rather than syntactic slot.
• I4. Each language divides up the range of be differently: Mandarin and Spanish carve identity, location, existence and temporary versus permanent attribution into separate copulas. That gives a ready semantic inventory for what English be is doing.
• I5. The paper separates "ungrammatical" from "infelicitous in context" and distinguishes verum focus (stressing the truth of the whole proposition) from VP-only focus, so judgments are graded rather than binary.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (tree): The NICE battery could serve as a deterministic check on whether a token labelled AUX or cop really behaves that way, flagging parser outputs or converted treebanks where seem/become got auxiliary treatment.
• I2 (frames): Frame building can take the predicate from the complement (e.g. a "doctor" attribute or an "in kitchen" location) and read tense and polarity off the copula, rather than inventing an empty "be" frame.
• I3 (world): When building entity states, be should normally yield a state or attribute, but an agentive reading (he is being silly) should yield an action event with an agent, which matters for event ordering and responsibility.
• I4 (back): Translation through the meta-world can store the specific relation (equative, locative, existential, temporary attribute) instead of a generic "be", then choose ser/estar/haber or shì/zài/yŏu correctly on output.
• I5 (learn): Negative controls in evaluation could include both starred and merely odd variants, e.g. *Does she be a doctor? versus a pragmatically off emphatic reply, to test whether models rank them appropriately rather than just reject both.