George Bernard Shaw is credited for saying, “England and America are two countries divided by a common language.” It is not clear whether or not he actually said that, but the sentiment is clear that people have a tendency to place judgement or discrimination on the basis of accent or dialect. Accent is the way a person sounds when speaking a given language. Even you, kind reader, have an accent. Your accent may be native to the area in which you live, or it may sound foreign based on how others talk in that region.
A classic (read: antiquated) example of accent discrimination can be seen in Shaw’s character Henry Higgins, dramatized and musicalized here:
Disclaimer: In our modern world, whether an accent sounds native or foreign to someone, it should never be the basis of discrimination. As researchers and lovers of language, we can appreciate accentedness for the beauty it brings to the language. A person should not be placed at fault for having an accent, but rather, we all should strive to be better listeners. *dismounts soap box*
There are many resources for listening to speakers of various accents and dialects. The Wildcat Corpus at Northwestern University has audio samples of native and foreign-accented speakers of English. The International Dialects of English Archive (IDEA) is the largest resource for listening to speakers of English across the world. Listening to these audio samples is a great way to understand what the speakers of various regions sound like.
Interestingly, visual information also colors our perception of accent. In a 2015 study, Babel & Russell used only native speakers of English, and found that Chinese Canadians were perceived as more accented when a picture of their faces were shown. In 2010, Hay and Drager found that stuffed toys influenced listener perceptions of New Zealand speakers: showing a stuffed kangaroo made the speakers sound Australian, and showing a stuffed kiwi made them sound like they were from New Zealand.
To help researchers better understand the audio-visual interplay in accent perception, Bailey McGuffin, Sara Incera, and Homer White (pictured above) have created a new corpus that provides audio and video samples of speakers who may or may not be perceived as having a “foreign” accent. The corpus was validated using listeners from the Bluegrass region of the state of Kentucky, so “foreign” was defined as
speakers who did not have a North American English accent.
The Bluegrass region makes up the northern part of Kentucky, and it is known for its bountiful bluegrass, calcium-rich soil, and abundant horse farms. (It remains to be seen if the horses of the region have a distinguishable accent.) Bluegrass is also a musical genre, and its popularity has spread far beyond this region. Here’s a taste of the modern Bluegrass style:
This region is also home to Eastern Kentucky University, where the new Bluegrass Corpus was developed. The corpus compiles a series of TED talk clips in an effort to provide audio-visual resources for researchers to investigate the role of visual information in accent perception. In their recent paper published in the Psychonomic Society’s journal Behavior Research Methods, McGuffin, Incera, and White explain how they created and validated the corpus.
The Bluegrass Corpus consists of 640 sentences spoken by 80 speakers with 8 sentences per speaker. Each excerpt is provided in three formats: text (.jpg), audio (.wav), and video (.mp4). The corpus is freely available through the Open Science Framework here.
The researchers selected the speech clips from the expansive TED database. They started with a total of 100 speakers, and they paired the speakers by individuals who looked alike (in terms of age, gender, height, weight, hair color, skin color, etc.). In each pair, one speaker had a native accent to North American English, and the other speaker had an accent of English outside of North America.
They validated the corpus in a series of three experiments. The experiments used three separate samples of undergraduate students from Eastern Kentucky University, and the students were asked to rate the excerpts on three qualities:
Experiment 1 – Accent: “In your opinion, how accented is this speaker?”
Method: Listeners heard audio clips of speakers saying a sentence.
Response: continuous scale (native to foreign)
Results: In 45 out of 50 pairs, listeners rated the foreign speaker as sounding more accented than the native speaker. Five pairs were rated as sounding equally accented, and these pairs were removed from the corpus.
Experiment 2 – Appearance: “In your opinion, how likely is this person to have a foreign accent?”
Method: Participants saw a still image of the speaker during their TED talk.
Response: continuous scale (not at all to very likely)
Results: In 46 out of 50 pairs, participants rated both speakers as equally likely to speak with an accent. Four pairs were rated as having different appearance, and these pairs were removed from the corpus.
Experiment 3 – Difficulty: “In your opinion, how difficult is this sentence?”
Method: For Experiment 3, the nine excluded pairs from Experiments 1 and 2 were removed, and one additional pair was removed to give an even 40 pairs in the final corpus. In this experiment, participants saw the written text of 12 sentences per speaker, and they rated the “difficulty” of each sentence. An easy sentence would have simpler ideas and would be understood by laypeople. A difficult sentence would have more complex concepts and could use jargon.
Response: continuous scale (easy to difficult)
Results: For each speaker pair, the researchers were able to identify 8 sentences from the native speaker that matched difficulty with 8 sentences from the foreign speaker.
The researchers provide more information on the corpus in a helpful walkthrough here. They also encourage users to update and expand the corpus by adding their own video selections, and to validate the corpus on other populations of listeners outside of the Bluegrass region.
We have seen that accent perception is influenced by the integration of auditory and visual information, and now the Bluegrass Corpus gives researchers a tool for further investigating this topic. As this area of language processing continues to grow, I hope it will lead to the Henry Higginses and Eliza Doolittles of the world to live together in peace and accented harmony.
Psychonomic Society article featured in this post:
McGuffin, B., Incera, S., & White, H. S. (2021). The Bluegrass corpus: Audio-visual stimuli to investigate foreign accents. Behavior Research Methods. https://doi.org/10.3758/s13428-021-01590-w