EnglishBlitz

TOEIC Link Grammar — Noun Clauses and Reported Speech: How Embedded Question Order, Tense Backshift, and Stance Verb Selection Drive Mid-Band Errors

Noun clauses and reported speech account for a disproportionate share of TOEIC Link grammar errors at the mid-band CEFR B1 to B2 range. The three error clusters — embedded question word order, tense backshift, and stance verb selection — are tightly correlated and respond to a single integrated drill protocol. This guide maps the underlying grammar, the four high-frequency error patterns, and the four-stage remediation sequence.

EnglishBlitz EditorialTeam·

TOEIC Link Grammar — Noun Clauses and Reported Speech: How Embedded Question Order, Tense Backshift, and Stance Verb Selection Drive Mid-Band Errors

Noun clauses and reported speech are syntactically connected but pedagogically separated in most TOEIC preparation materials, and the separation is responsible for a substantial share of the mid-band grammar errors observed at CEFR B1 and B2. Candidates learn the embedded-question word-order rule in one lesson, the tense-backshift rule in another, and the reporting-verb selection rules in a third — and the integrated structure (a stance verb taking a noun-clause complement that requires backshift and embedded-question order) is rarely drilled as a unified construction. The result is a predictable error profile in which candidates produce correct individual elements but combine them incorrectly under timed conditions.

This guide describes the underlying syntactic structure shared by noun clauses and reported speech, maps the four high-frequency error patterns that drive mid-band losses on TOEIC Link grammar items, and proposes a four-stage drill sequence that integrates the three rules into a single fluent production system. For broader context on the grammar module, see the guides on grammar verb tenses, grammar relative clauses, and grammar passive voice and causative.

Why noun clauses and reported speech belong in the same lesson

A noun clause is a finite clause that functions as a noun in a larger sentence — as the subject of a main verb, as the direct object of a verb, as the complement of a preposition, or as a predicate complement after a copular verb. Reported speech is a specialized application of the same structure in which the main-clause verb is a verb of saying, thinking, or perceiving and the noun clause reports the content of the speech or thought.

The two structures share three syntactic properties.

Shared property 1 — the noun clause begins with a complementizer. The complementizer is "that" (for declaratives), or a wh-word (for questions). The complementizer "that" is optional in object position after high-frequency reporting verbs ("I think (that) she's right") but obligatory in subject position ("That she was right was obvious"). Wh-complementizers are obligatory in all positions.

Shared property 2 — the noun clause has its own subject and finite verb. The embedded subject does not move out of the clause, and the embedded verb agrees with the embedded subject independently of the main-clause verb.

Shared property 3 — the noun clause has declarative word order regardless of the meaning. A direct question has subject-auxiliary inversion ("Where is she?"), but the same content embedded as a noun clause has declarative word order with no inversion ("I want to know where she is"). The declarative-order rule applies to both pure noun-clause uses and reported-speech uses.

Because the three properties are shared, an item that tests one property frequently tests another. A TOEIC Link grammar item that targets reported speech often also targets embedded-question word order; an item that targets stance verb selection often also targets tense backshift. The mid-band candidate who has learned each rule in isolation fails to apply all three correctly under time pressure.

The four high-frequency error patterns

Pattern 1 — embedded question inversion error

The most common error in the noun-clause category is the failure to remove subject-auxiliary inversion when a direct question is embedded as a noun clause. The candidate produces:

  • *"Could you tell me where is the meeting room?"
  • *"I wonder why did she leave early."
  • *"Do you know what time does the conference start?"

The correct forms are:

  • "Could you tell me where the meeting room is?"
  • "I wonder why she left early."
  • "Do you know what time the conference starts?"

The rule is straightforward — declarative word order in the embedded clause — but the production error is persistent because the candidate has internalized the direct-question form as the canonical interrogative pattern. The remediation requires sustained re-drilling of the embedded form in production tasks rather than recognition tasks.

Pattern 2 — tense backshift over-application

After a past-tense reporting verb, the embedded verb typically shifts one tense backward — present becomes past, past becomes past perfect, will becomes would, can becomes could, may becomes might. The rule is well known, but TOEIC Link candidates frequently over-apply it, producing:

  • *"He said that the office building had been built in 1995." (correct only if the reporting context implies the relevant time was even earlier; usually "was built in 1995" is correct because the statement reports an enduring historical fact)
  • *"She mentioned that water boils at one hundred degrees." (correct as written — universal truths do not backshift, but candidates over-shift to "boiled")
  • *"He told me that the company had been founded in Boston." (over-shifted from the appropriate "was founded in Boston" when the founding is a stable historical fact)

The remediation is to teach the three exceptions to backshift explicitly: universal truths, currently-true facts presented as still currently true, and recent past events. A candidate who applies backshift mechanically will lose points on items that test the exceptions.

Pattern 3 — stance verb selection mismatch

Stance verbs (verbs of saying, thinking, perceiving, knowing, doubting, suggesting) impose different complementation patterns. TOEIC Link grammar items exploit these differences.

Verbs taking a that-clause complement. Say, think, believe, know, realize, hope, expect, mention, announce, claim. These verbs can take a finite that-clause: "He said that the meeting was canceled."

Verbs taking a to-infinitive complement with optional object raising. Want, expect, believe, ask, tell, advise, instruct. These verbs typically take an object plus to-infinitive: "I asked him to send the report." The pattern is not a noun clause; it is an object-control construction. Candidates who treat these verbs as noun-clause heads produce errors.

Verbs taking a gerund or bare-clause complement. Suggest, recommend, propose, insist. These verbs frequently take a subjunctive or bare-form clause: "He suggested that she leave early" (subjunctive) or "He suggested leaving early" (gerund). The form *"He suggested that she leaves early" is non-standard in formal registers and is the target of mid-band test items.

Verbs taking a noun-clause complement with mandatory "that" retention. Argue, contend, hypothesize, predict, anticipate. These verbs require "that" complementizer retention and resist the deletion that is permitted with "think" and "believe."

Mid-band candidates frequently substitute one stance verb for another and produce complementation mismatches. TOEIC Link grammar items test this systematically by presenting four stance verb options with otherwise identical structure and asking the candidate to identify the one verb that matches the complementation pattern of the rest of the sentence.

Pattern 4 — time and place adjustment failure in reported speech

When direct speech is converted to reported speech, time-and-place deictics typically shift to align with the new reference point: "today" becomes "that day," "tomorrow" becomes "the next day" or "the following day," "here" becomes "there," "this" becomes "that." Candidates who learn the tense-backshift rule but not the deictic-adjustment rule produce:

  • *"He said yesterday that he will see me tomorrow." (mixed reference frames — should be "the next day" or rephrase)
  • *"She told me last week that this is a difficult problem." (should be "that was" or "that is" — the simple-present form requires the proposition to be currently true at the reporting moment)

The remediation is to drill the deictic-adjustment rule alongside the backshift rule rather than treating them as separate items.

The four-stage drill protocol

A high-yield drill protocol integrates the three rules — embedded-question word order, tense backshift, and stance verb selection — into a single fluent production sequence.

Stage 1 — recognition discrimination. The candidate is shown twenty pairs of sentences, one correct and one incorrect, and identifies which is correct. The error in the incorrect sentence rotates among the four patterns above. The goal is to build conscious awareness of each error type.

Stage 2 — transformation production. The candidate is shown twenty direct-speech sentences and produces the reported-speech equivalent, applying backshift, deictic adjustment, and embedded-question order. The goal is to integrate the three rules in a controlled production task.

Stage 3 — stance verb substitution. The candidate is shown twenty sentences with a stance verb in place and substitutes a near-synonym from a constrained list, adjusting the complementation as needed. The goal is to expose the stance-verb / complementation pairing so the candidate selects the right verb when the test item presents alternatives.

Stage 4 — timed integrated production. The candidate produces five reported-speech sentences from prompts within sixty seconds each. The goal is to compress the integrated three-rule production into the timing range required by the test.

Candidates at CEFR B1 entering this protocol typically show a measurable improvement in noun-clause and reported-speech accuracy on TOEIC Link practice items within four weeks. The persistent error patterns shift from the inversion error and the over-backshift error (which respond fastest to remediation) to the stance-verb selection error (which responds more slowly because it depends on lexical exposure).

Implications for the Reading and Listening modules

The noun-clause and reported-speech construction is not exclusive to the Grammar module — it permeates the Reading module (where reported-speech embedding distinguishes attributed claims from direct claims and is the target of paraphrase-recognition items) and the Listening module (where speakers report colleagues' positions, customer feedback, and prior decisions through reported speech). A candidate who masters the construction in the Grammar module also gains comprehension reliability in Reading and Listening passages that embed the construction.

For complementary drilling on the Reading and Listening implications, see reading paraphrase recognition techniques and listening inference and implication questions.