Assessing workplace English means measuring whether a person can complete the communication tasks a job actually contains — understanding a customer, writing a clear escalation note, explaining a fault to a colleague — at a defined level of accuracy. It measures intelligibility and task completion. It does not measure accent, vocabulary size, or how closely someone resembles a native speaker.
That distinction sounds pedantic. It is the difference between a screen that predicts job performance and one that quietly filters on social class and mother tongue. Most English screening in Indian hiring — the 1-to-5 "communication" box on the interview form, the "MTI check", the group discussion round — fails it. Here are five beliefs that keep the practice in place, and what the evidence says about each.
Myth 1: A candidate's accent tells you how well they will communicate
Accent, comprehensibility (how much effort a listener spends) and intelligibility (how much the listener actually understands) are three separate things. Munro and Derwing's long line of work, revisited in their 2020 "redux" paper, finds the three are related but partially independent, and that speech can be heavily accented while remaining highly intelligible. Rating accent is not a shortcut to rating understanding; it is a different measurement.
Language assessment has already acted on this. When the Council of Europe revised the CEFR's phonological control scale for the 2018 Companion Volume, it rebuilt the scale around intelligibility and removed native-speakerism, on the reasoning that "intelligibility is far more important to communication than accent". The old 2001 scale, the Council notes, implied that progress meant sounding more and more like a native speaker.
Meanwhile the cost of rating accent is measurable. A meta-analysis in Personality and Social Psychology Bulletin pooling 139 effect sizes from 4,576 participants found candidates with standard accents were rated more hireable than candidates with non-standard accents at d = 0.47. A separate 2025 meta-analysis in the International Journal of Selection and Assessment, covering 120 studies and 20,873 participants, put the same effect at d = 0.46.
The detail that should end the argument: in both meta-analyses, comprehensibility ratings did not significantly moderate the bias. Interviewers were not penalising candidates they struggled to understand. They were penalising candidates who sounded different. The 2025 analysis also found the effect was not reduced by interview modality — moving to video does not wash it out.
Myth 2: The national proficiency ranking tells you about your applicant pool
India scored 484 on the 2025 EF English Proficiency Index, ranking 74th, six points down on the previous edition and below the global average of 488. Within that, reading sat at 494 and writing at 504, while listening trailed at 457.
Useful as a directional signal. Useless as a screening input, and EF says so itself: the index draws on 2.2 million test takers who chose to take a free online test, a population EF describes as "self-selected and not guaranteed to be representative", skewed young (median age 26, 85% under 35) and structurally excluding anyone without reliable internet.
A national index cannot tell you what proportion of your applicants can handle your calls. Only a job-anchored measurement on your own pipeline can, and the gap between the two is exactly where hiring teams substitute assumption for data.
Myth 3: The "communication: 3/5" box is a measurement
A holistic impression score has no defined construct behind it. Two interviewers marking 3 may mean fluency and grammatical accuracy; a 3 from a panel that just heard a strong regional accent may mean something else entirely. Nothing in the process makes those numbers comparable, and nothing records what evidence produced them.
The fix is not a longer form. It is splitting the judgment into rows a rater can defend:
- Task completion — did the candidate get the required information across, in full?
- Intelligibility — what proportion of the response was understood on first hearing?
- Accuracy where it matters — numbers, names, dates, product terms, amounts
- Register and repair — can the candidate check understanding, apologise, clarify, escalate?
No row is called "accent" and no row is called "MTI", because neither is a job requirement. Raters then need anchored examples and periodic calibration — untrained raters drift, and a rating scale is only as good as the shared standard behind it.
Myth 4: Machine scoring of spoken English either solves this or cannot be trusted
Neither. A 2025 study in PLOS One compared three automated speaking-assessment systems against three experienced human raters on an IELTS-adapted speaking test. The human raters agreed well with each other (ICC 0.713 for a single rater, 0.882 averaged). Two of the three systems correlated strongly with the human average — r = 0.854 and r = 0.873 — with no significant difference in mean score.
The third is the cautionary half. It correlated just as strongly, r = 0.869, yet ran systematically high: a mean of 92.00 against the human average of 86.43, a difference that is enormous in effect-size terms. Correlation was fine; calibration was not. If you check only the correlation, you will pass a system that inflates every candidate's score by roughly six points — which changes who clears your cut score without changing the rank order at all.
The study also found automated systems struggled with pragmatic and sociolinguistic appropriateness — the part of workplace English that decides whether a refusal sounds firm or rude. Two implications follow. Check agreement and mean difference before you trust a scoring engine. And treat a 30-speaker study as a reason to run your own agreement check on your own population, not as a generic warranty. This is why AssessAll reports AI-graded scenario responses alongside integrity bands rather than as a bare number: a score is a claim, and a claim needs its evidence attached.
Myth 5: Setting the English bar high is the safe option
It is the opposite of safe. The US EEOC's enforcement guidance on national origin discrimination treats an accent-based employment decision as lawful only where the accent "interferes materially with job performance", and holds that an English fluency requirement is "permissible only if required for the effective performance of the position". Requiring more fluency than a role needs, or applying one standard across dissimilar roles, is itself evidence of a problem.
Most Indian employers are not litigating under Title VII. The discipline still applies, because an over-specified English bar does real damage at home: it shrinks tier-2 and tier-3 pipelines, rejects candidates who would have performed, and converts a hiring criterion into a proxy for schooling and background. The defensible position is the narrow one — this role requires this level on these tasks, and here is the evidence.
What to do on Monday
- Write the communication task inventory for one role: every task, its channel, and who the counterpart is.
- Set a required level per task, not per person, using CEFR descriptors as anchors.
- Build the assessment from job-shaped tasks — listen to a recorded customer call and write the ticket note; handle a chat escalation; explain a fault in 60 seconds.
- Score task completion and intelligibility as separate rows. Delete anything that scores accent.
- Validate your scoring engine or raters on your own candidates: report correlation and mean difference.
- Set the cut score from the job's demands, not from a habitual 70%.
- Run an adverse-impact check by region and first language before launch, and again at quarter end.
The teams that get this right stop asking whether a candidate "has good communication" and start asking whether they can do the three things the job requires, in the medium the job uses. That question has an answer you can measure, defend, and record — on a Skill Passport or anywhere else — and it does not depend on how anyone sounds.
Takeaway: Accent is not proficiency, and a national ranking is not your applicant pool. Measure the communication tasks the role actually contains, score intelligibility rather than accent, and check your raters and your scoring engine for calibration as well as agreement.