Why you shouldn’t buy books to help you with your pronunciation

There are lots of books out there that say they can help you learn good pronunciation. But can you actually learn pronunciation from reading about it? On the one hand, it seems intuitive that to learn good pronunciation you need a teacher who can hear how you are speaking now and who can tell you exactly what you need to change and how to change it. Adding to this, a book can’t let you hear the correct pronunciation or let you see how your mouth should be moving. On the other hand, if you don’t have a teacher, then a book is better than nothing, right?

So can written instructions help people to improve their pronunciation? I haven’t been able to find any studies that looked at this, so I decided to do one myself. I needed a specific pronunciation issue to tackle, and as I am an English teacher for Dutch people, I knew immediately which issue I wanted to address!

Dutch people do this thing…

Dutch people do this thing where words which end in voiced consonants like /d/, /g/, and /z/ will sound like their unvoiced counterparts /t/, /k/, and /s/. In practice, this means “bed” sounds like “bet”, “dog” sounds like “dock”, and “eyes” sounds like “ice”. It’s commonly known as final devoicing, though this suggest it is all about the sound of that last consonant. You’d think that, but it’s actually the vowel length that’s much more important.

I have been fascinated by final devoicing ever since a Dutch talent-show contestant called Dewi released a single in 2003 featuring the lyric “being here with you is paradise, I can see forever in your eyes”. For me, as an English native speaker, “paradise” and “eyes” do not rhyme.

Final devoicing also happens to Dutch people with otherwise impeccable English. One of the Dutch presenters of the Eurovision Song Contest gave me a perfect example to prove my point just a few weeks ago; Chantal Janzen speaks excellent English with an accent modelled on RP (“queen’s English”). But at 2:55 in the final she nonetheless asked permission to “move to the first face of the results”.

It’s not only Dutch people, by the way, Germans do it too, as do most speakers of Slavic languages such as Russian and Polish. Wikipedia has a good list.

I made Dutch people say “eyes” and asked English speakers if they heard “ice”

Glass, Eye, Drink, Weird, Creepy

In May 2021 I asked 73 Dutch people to record five words: “book, eyes, mango, bed, label”. I then presented them with some written pronunciation tips, and asked them to record the same five words again.

24 people were in a control group, and the only tip they read was that it was a good idea to articulate properly, and that they shouldn’t be ashamed of their accent (most Dutch people hate the way they speak English, they think the Dutch English accent is ugly and unrefined, an idea perpetuated by the media. But actually, research has shown that English speakers think the Dutch English accent is fine.)

The experimental groups got a specific tip on how to pronounce words like “eyes” and “bed”, which many Dutch speakers will pronounce like “ice” and “bet”. These instructions can be found here (in Dutch). The main point made is that the trick is not to concentrate on making the /d/ in “bed” more voiced (in other words more like a /d/ and less like a /t/), but instead to make the vowel longer: “beeed”. This helps much more to make it sound more like “bed” and less like “bet”. The instructions were almost exactly the same for the two groups, except the first group got the instructions with audio examples, and the second group got the instructions in writing only.

Judging the experiment: playing the recordings to English speakers

I then put all 146 recordings (two per speaker) of those Dutch people saying “book, eyes, mango, bed, label” into a questionnaire which was filled in by 411 English speakers (not English native speakers per se; people from any country who spoke English). The software randomly selected six of the recordings per participant, and asked them to select what they heard, and how sure they were. The question looked like this:

One remark I got from some of my respondents, was that by giving multiple choice options between “ice” and “eyes” and “bet” and “bed”, I was leading the question.

And the answer is that indeed I was! The thing is, I wasn’t looking to answer the question “how well did this Dutch person succeed in saying eyes“, I was looking to answer the question “how much better did this Dutch person succeed in saying eyes after the intervention.” Because I wanted to be able to compare, I purposely allowed my judges to think that the person might have been saying ice. Firstly by making the answer options multiple choice, secondly by starting the answer options with ice. This way, if the word the Dutch person was saying was even a little bit close to ice, I’d be able to see it in my results.

Another remark I got was that the sequence of five words was too unnatural.

I agree with this. The ideal way of running this experiment would have been to get my Dutch English speakers to say a sentence like “She has green eyes,” then cut only the word “eyes” from the sentence and present that to my judges. And then do the same with “bed”. But this would have meant cutting up 292 audio files, and while I was designing my study I felt this was not a feasible workload considering I had a deadline. In hindsight, I have to say that the part of the study where I was fiddling with audio files was actually less work than I had thought, so perhaps I should have done this after all.

The sequence of words (“book, eyes, mango, bed, label”) might seem really random, but it was actually chosen very carefully. “Book”, “mango” and “label” are words that present no real pronunciation problems for Dutch people; I didn’t want them thinking too much about words that didn’t matter (the speakers did not know that it was only “eyes” and “bed” that I was interested in). All five words are nouns, short, well-known even for Dutch people with a low level of English, and none of them have confusing spelling such as, to name an example that haunted my teenage years, “draught” (for the non-native speakers reading this; it’s pronounced “draft” and, yes, English spelling is spectacularly awful).

The first sounds of “mango” and “label” (sonorants, for the phonetically inclined) interfere the least with the “s” of eyes and the “d” of bed; if the words had been “eyes, cat” then the voicelessness of the /k/ sound would have bled over into the last sound of “eyes”, making it potentially sound more like “ice”, and if the words had been “eyes, bottle” the inverse would have happened.

Then there were also the judges to keep in mind; had I chosen five words such as “water, eyes, sleep, bed, label” then the “water” might have made people more prone to choose “ice” over “eyes” and the “sleep” might have made people more prone to choose “bed” over “bet”. So I made sure to choose words that had nothing to do with eyes, ice, beds or bets.

What were the results?

I had only one statistically significant result, and it was this: If you tell people to mind their pronunciation, then their pronunciation will improve. A bit.

But, importantly, I did not find a difference between my control group and my experimental groups, meaning the specific instructions to combat final devoicing had no effect.

I won’t go into the details of my statistics, but how it worked was basically this: The choices of my judges were transformed to scores: “It’s definite ice” became 1, “I’m pretty sure it’s ice” became 2, “I think it’s ice” became 3, “I think it’s eyes” became 4, “I’m pretty sure it’s eyes” became 5 and “It’s definitely eyes” became 6. That way I could work out averages. If a certain speaker was rated on average as, say 4.6, then that speaker did a reasonably good job at pronouncing “eyes” like “eyes” and not like “ice”, but it could have been better. Then I averaged all the average scores for all the speakers in one group, so I could compare the groups. SPSS is a program that can tell you if the averages are different enough to be real, and not just a coincidence.

As you can see, there’s hardly any difference between the recordings that people got before the instructions (time 1) and the ones after (time 2) – it’s still just big enough to be statistically significant, though. The difference between conditions is not statistically significant.

Why didn’t my written instructions work?

I had a strong suspicion that I would see these results the moment I started listening to the audio recordings I had gathered, especially for “eyes”. Why? Because I had an awful lot of Dutch people whose pronunciation of “eyes” and “bed” was absolutely fine in the first recording. When I went to listen to the second recording of these near-native speakers, for quite a few of them I heard a deterioration of the pronunciation; they were now thinking about it, trying to do a good job, trying to make the vowel longer, but in trying to make the vowel longer, also making the /s/ longer, so they ended up saying “eeeeeeeyssssssss”. Which sounds more like “ice” than the way they were saying it before. And indeed when I looked at the numbers I could see that lots of people who got a good score in the first recording got a worse score in the second. Not the majority, but a good chunk.

“Bed” had less of this problem, as it is not possible to make the sound /d/ any longer than it already is. But something about concentrating on the pronunciation of the word made some people pronounce it less clearly nonetheless.

The moral of the story

The main problem with written pronunciation instructions is that they cannot work if the person trying to improve their pronunciation doesn’t know 1) if the issue that is being discussed is in fact an issue for them, and 2) if they are applying the instructions correctly.

To check this idea I looked at only those respondents who got a bad “grade” in the first recording. Doing this is statistically extremely dubious, because I was left with very, very small groups (between 9 and 14 speakers per group), not to mention this is a classic example of P-hacking. But I can look at a “trend” (a popular word among students-without-significant-results the world over). For “bed”, especially, the graph looks much more like you would expect if you only give instructions to people who actually need them.

Crucially, these people move from “I think it’s bet” (3) to “I think it’s bed” (4) which was the whole point of the pronunciation lesson.

Conclusion: it’s all up to teachers

In my study, I attempted to give a one-size-fits-all pronunciation lesson in writing. But when I listened to my audio recordings what I mostly wanted to do was crawl through the internet cable and coach the speakers personally, because they were all doing slightly different things, and had I been there in person I would have been able to say “you don’t need this tip, your pronunciation was fine!” or “Your “bed” sounds like “bad”, let’s work on that first.” etc.

I also keep thinking of the English teacher who made a great remark in my pre-study: She said that it was already a challenge to get her young high school students to speak English at all, and that correcting their pronunciation would just make them even more intimidated and unwilling to practice speaking. I think she was absolutely right; there is a time and place for pronunciation training, and it is up to the teacher to decide if the student is ready for it.

I think pronunciation tips on paper can work, but there needs to be a person to guide the learner, tell them which tips they need, which ones they don’t, and monitor their progress.

I also now think teacher training is very important. When I did my master’s in Utrecht to become an English teacher, giving pronunciation training was not addressed at all. So I see some room for improvement there.

And the conclusion of the conclusion:

Dear English teachers: your work matters.

Note: This article is a re-working of a much longer previous version that I published on this website just after I got my statistical results but before I had put my thesis together. It was way too long and very rambling.

Note 2: I got 85% for my thesis, yay! I’m not totally happy with it (is anyone ever?); it’s an interdisciplinary thesis and I found the tightrope between psychology, linguistics and educational science a difficult one to walk. But if you are curious, you can download it here.

Heddwen Newton is a teacher and translator. Her website EnglishforDutchpeople.nl is about efficient and unconventional ways for Dutch people to improve their already good English, and other nerdy stuff to do with English and Dutch. She also owns the Dutch website HoezegjeinhetEngels.nl where she discusses difficult-to-translate Dutch words and their least-bad English translations.

Leave a Reply

Your email address will not be published.