Many Japanese adults think they can never hear the difference between L and R. But that is not true. In the 1990s, researchers in the US trained Japanese adults.

The adults listened to many words read by many different people. They answered "L" or "R" and checked the answer right away. Their listening got better. Six months later, they still did better than before training. Their pronunciation also improved, even without speaking practice.

Studies show that Japanese babies at six to eight months can tell English L and R apart as well as American babies, but this ability declines by about their first birthday. Can adults learn it again? In 1991, researchers at Indiana University trained Japanese adults to identify L and R in many words spoken by several different speakers.

After each answer, the learners were told immediately whether they were right. Their accuracy improved, and the skill even worked with new words they had not practiced. This method is called high variability phonetic training. Later reports showed that most of the gains were still there three and six months later, and that the learners' pronunciation of L and R improved, although they had only practiced listening.

The English /r/-/l/ contrast is notoriously difficult for Japanese speakers, and many assume adults cannot master it. Developmental research explains the difficulty: Japanese and American infants discriminate the two sounds equally well at six to eight months, but by ten to twelve months Japanese infants have declined as their perception tunes to Japanese, which does not use the contrast. A landmark series of studies beginning in 1991 showed that this tuning is not permanent.

Logan, Lively and Pisoni trained Japanese adults to identify /r/ and /l/ in natural recordings of many words, produced by multiple speakers, with immediate feedback after each response. Accuracy rose and generalized to untrained words and voices. This approach, now called high variability phonetic training, is thought to work because exposure to many voices helps learners extract the features that remain constant across speakers.

However, each study involved few participants, and the 1991 study found only small but reliable gains. Follow-up reports found that much of the improvement was retained three and six months later, and that participants' pronunciation improved even though they had practiced only listening. A 2018 meta-analysis of 25 years of research confirmed that perception training tends to transfer to production.