Your Accent Isn't the Problem: What Actually Makes You Hard to Understand

You have spent years trying to sound less Spanish in English. You record yourself, you hear the accent, and you assume that's why people ask you to repeat. It isn't. How foreign you sound and how much people understand you are two separate things, measured separately, and they barely move together: in the most complete meta-analysis to date, 49 studies published between 1995 and 2023 covering 141 effect sizes, the link between accent and actual understanding is the weakest of every relationship examined. What decides whether people understand you is a short list of specific features, and "accent" in general is not on it.

Accent, effort, and understanding are three different things#

Pronunciation research splits into three measures what everyday conversation lumps into one. Murray Munro and Tracey Derwing defined them in 1995, and the field has used their vocabulary ever since: accentedness is how far your speech sits from the local variety, comprehensibility is how much effort a listener spends following you, and intelligibility is how much they actually understand. The first two are rated on scales; the third is scored by dictation and comprehension questions, because it measures an outcome, not an impression.

The three-way split is the finding, not academic hair-splitting. The two researchers who coined the terms put it bluntly in their own conference material: "A speaker can have a very strong accent, yet be perfectly understood." In their 1995 study, native listeners transcribed sentences from Mandarin speakers. Plenty of utterances came back word for word — fully understood — and still scored at the heavy end of the accent scale.

The weakest correlation in the literature#

The correlation between how accented you sound and how much people actually understand is r = .32. Tuc Chau and Amanda Huensch, writing in Studies in Second Language Acquisition, calculated the mean correlations among the four dimensions of L2 speech; that one landed lowest of the five, a small effect by the field's own benchmarks. Their conclusion is the line to frame: "having an accent does not mean being unintelligible."

Accent does track listener effort closely, at r = .75. Sounding foreign costs the person listening some work, but it doesn't stop them understanding you. That's the difference between a road with potholes and a road that's closed. Chasing a zero accent attacks the wrong half of the problem.

The features that actually decide whether you're understood#

John Levis named the two competing philosophies of pronunciation teaching in a 2005 paper that, by his own count, has been cited more than 800 times: the nativeness principle and the intelligibility principle. Fifteen years later he shut the door on the first one: "speakers who are perceived as strongly accented can also be highly intelligible." Under the intelligibility principle, accentedness is largely irrelevant; what counts is whether the message lands.

Not every sound error costs the same. The variable that ranks them is functional load: how many word pairs a contrast actually distinguishes. A review in Frontiers in Communication concludes that high-functional-load errors are the only ones tied to a measurable drop in comprehensibility, naming /v/ pronounced as /b/ as its example. That's the Spanish speaker's case exactly: very and berry. Missing the th in think almost never breaks anything.

Where the emphasis falls matters more than any single vowel does. Laura Hahn played the same university lecture three ways, with the primary stress correctly placed, misplaced, or missing. With the stress in the right place, students recalled significantly more content and rated the lecturer more favorably, on identical wording.

As a Spanish speaker, your three real weak spots are measurable. First, the long–short vowel contrast: Spanish has one /i/ where English has two, which is why sheep and ship keep collapsing into one word. A study acoustically tracked 40 Spanish-speaking postgraduates through a year at a British university to see whether they pulled the pair apart. Second, rhythm: English reduces unstressed syllables to schwa and Spanish doesn't, and in a study of 60 speakers in Miami, age of English learning explained 51% of the variation in how long those reduced vowels ran. Third, word-initial consonant clusters, where the extra e in estrong sneaks in.

Part of the burden belongs to the listener#

Some of what you read as "they don't understand me" never came out of your mouth. In a 1992 experiment, Donald Rubin had American students listen to a native speaker's recording while looking at a photograph. When the photo showed an Asian face, students reported hearing an accent that wasn't there and scored worse on the comprehension test. Same audio.

Expectation operates on judgment, not on hearing. Yi Zheng and Arthur Samuel replicated the effect with still photos, but the shift largely vanished once they swapped in dubbed video. Their conclusion: visual information affects "the interpretation, but not the perception, of accented speech." Rosina Lippi-Green has spent decades naming where that leads. Communication breaks down less because of accent than because of "negative social evaluation of the accent in question, and a rejection of the communicative burden". Understanding someone is a two-person job. An accent, as Munro puts it, is not a disorder; it's normal variation in human speech.

What practice actually changes#

Exposure alone plateaus early, which is bad news for anyone counting on immersion. Munro, Derwing, and Kazuya Saito followed 24 adult immigrants for ten years, measuring a single sound: word-initial /p/. The improvement clustered in the first months, while they were still in class, and between year 1 and year 7 there was no significant difference. Living in the country isn't the variable; instruction is.

Targeted instruction works, as long as you measure it where it shows. A meta-analysis of 77 studies published between 1982 and 2017 found clear gains feature by feature, one consonant or vowel contrast or stress pattern at a time, and much murkier ones when the outcome was "how accented does this person sound." Train small, measurable targets instead of a whole sonic identity.

The short list already exists. The British Council publishes Jennifer Jenkins's core for international English: every consonant except the th sounds, consonant clusters at the start and middle of words, the contrast between long and short vowels, and nuclear sentence stress. Jenkins adds the argument that closes the loop: "native speaker accents are not necessarily the most intelligible or appropriate accents." One honesty check: that core is built for conversations between non-native speakers, the majority group at roughly 1.5 billion, and it leaves out word stress, which Hahn's data flag as decisive in front of native listeners. Pick according to who you're talking to.

None of this counts until it comes out in real time instead of on a worksheet. Could you say "a sheet of paper" out loud right now, without thinking, and keep it from landing as shit? Closing that gap between knowing a contrast and producing it under time pressure is what Tabla drills: you say the phrase out loud against the clock, and the app scores your pronunciation on the spot.

Stop chasing a neutral accent nobody asked you for and go after the four or five features that genuinely change what the other person understands. It's a shorter list, and a far less demoralizing one.

Try this today: record yourself saying "I didn't say he stole the money", stressing a different word each time, and listen to how the meaning shifts on every take.