The Effect of Dialect Features on the Perception of “Correctness” in English-Word Voting
Patterns on Forvo.com
Jessica Grieser (Georgetown University)
Forvo.com is a user-driven online dictionary of word and short phrase pronunciations, where individuals may record
pronunciations and rate those of others on their “correctness.” Launched in January 2008, it archives over 188,000 pronunciations
in 217 languages as of May 2009. This paper examines the ratings of pronunciations from speakers in the United States, England,
and Australia to determine the factors most responsible for high- and low-scoring English pronunciations. Niedzielski (1999)
found that perceived speaker locale affected naïve listener perception of phonetic variables. This paper exams two variables
which, in combination with listeners’ perception of speaker locale, affect the “correctness” rating of English pronunciations on
Forvo: the perception of hypercorrection as evidenced by the realization of intervocalic /t/, and the link between perceived
speaker locale and topic of the word being pronounced. Released intervocalic /t/ is a well-documented feature of British and
Australian English (Wardhaugh 1999, Wolfram and Fasold 1974, Bayard et. al. 2001). Within the sample of 187 pronunciations
used for this data, only released-/t/ pronunciations by British and Australian speakers received average scores in the high range
(greater than 4.0 on a 5-point scale), suggesting that Forvo voters consider released /t/ a hypercorrect feature when from a US
English speaker. Voters also show a strong preference for dialect features to match the topic of the word or phrase being
pronounced. Listeners prefer hearing US locations or personalities pronounced by a US speaker and vice versa, as evidenced by
the lack of any high-scoring pronunciations of words by speakers whose dialect locale did not match the topic of the pronounced
word. Both of these patterns suggest that naïve listeners attend extensively to dialect when making judgments about the overall
correctness of features in even single-word pronunciations.
Method:
Data were gathered for this study over a period of six weeks by monitoring the website for changes in the
most frequently played pronunciations (which therefore are the most likely to have received higher
numbers of votes) and then examining the profiles of those speakers for number of votes and other
pronunciations. Pronunciations selected were limited to those that had received three votes or more, and
then analyzed based on their average score. The score descriptions used by Forvo.com are as follows: 1worthless, 2-no native or bad sound, 3-good, 4-great, 5-perfect!The collected pronunciations were then
grouped into low-scoring (1-1.9) medium-scoring (2.0-3.9) and high-scoring (4.0+) pronunciations and
analyzed for features which affected this score.
Two primary factors seemed to influence voters’ perceptions of the correctness of a pronunciation most
heavily: perceived hypercorrection, and topic sensitivity.
Medial /t/ as evidence of hypercorrection
Score range
# of Tokens
Realized as /ɾ/,
US
Realized as /t/
US
Realized as
/t/ Non-US
3
Realized as
/ɾ/
Non-US1
0
Low (<1.9)
Medium (2.03.9)
High (>4.0)
0
11
2
2
7
4
0
0
3
In Standard American English, we can expect a flapped variant of /t/ to be realized
intervocalically, and in British and Australian English, we can expect /t/ to be realized as /t/ (Bayard, et.
al.) The Forvo voters seem to mirror this expectation, voting U.S. speakers’ pronunciations high only
when the /t/ is realized with the flapped variant and voting non-US speakers’ high only with the released
variant. This points toward the probability of the voters’ attending to dialect features and perceiving the
1
None of the 18 tokens with medial /t/ were spoken by the two Canadian speakers.
released /t/ variant as hypercorrect (and therefore a poor pronunciation) when it is produced by a U.S.
speaker.
Topic Sensitivity and Dialect Locale
Table 4.2.1 Topic Sensitivity: U.S. Topics
Score range
Number of Tokens
Low (<1.9)
1
Medium (2.0-3.9)
10
High (>4.0)
8
U.S. Speaker
0
3
7
Non-U.S. Speaker
1
7
0
Table 4.2.2 Topic Sensitivity: Non-U.S. Topics
Score range
Number of Tokens U.S. Speaker
Low (<1.9)
3
3
Medium (2.0-3.9)
4
2
High (>4.0)
1
0
Non-U.S. Speaker
0
2
1
For this part of the study, 27 tokens were selected from the total set whose topics were readily
identifiable as particularly U.S. or non-U.S. in nature. Words considered as identifiably “US” or “non-US”
in topic included locations, such as New York vs. London, or celebrities, like Tom Cruise vs. David
Beckham. The absence of speakers in one locale or the other for both the high and low scoring tokens as
indicated in tables 3 and 4 is evidence that where the topic is U.S. in nature, an American accent is
preferred for the word’s pronunciation, and the opposite is true where the topic is notably British or
Australian. The contrast between the two sets of middle scores, while still supporting the conclusion that
dialect locale preference is dependent on topic, seems also to continue to support the apparent general
bias on Forvo towards Standard American accents with the non-US speakers being dispreferred for US
topics at a greater than 2:1 ratio versus the 1:1 ratio for non-US topics.
Preliminary Conclusions
This preliminary study makes evident that Forvo voters are sensitive to issues of dialect in their
evaluations of “correctness” of a given English pronunciation. They attend to dialect features both at the
micro level, when evaluating the significance of a released or flapped medial /t/, and also at the macro
level by preferring dialect that matches the locale associated with the topic of the word being pronounced.
Although the large number of US-based voters has not resulted in equally large numbers of negative votes
for non-American pronunciations, a slight overall bias toward American English is evident when
examining more subtle distinctions, especially among middle-scoring pronunciations.
Further study of English pronunciations within Forvo might focus more heavily on topic
sensitivity, particularly as the sites’ usership continues to increase and more and more words are added
and tagged as relevant to U.S., U.K., or Australasian culture. In addition, as more pronunciations are
added to the database, more words will inevitably be associated with multiple pronunciations, and the
differences in scores for variants of the same word would provide some of the experimental control that
is difficult to obtain at this time.
References:
Bayard, D., A. Weatherall, C. Gallois, and J. Pittam. 2001. Pax Americana? Accent evaluations in New
Zealand, Australia and America. Journal of Sociolinguistics 5, no. 1.
Clopper, C. G., and D. B. Pisoni. 2004. Some acoustic cues for the perceptual categorization of American
English regional dialects. Journal of Phonetics 32, no. 1: 111-140.
Forvo.com www.forvo.com.
Giles, H. 1970. Evaluative Reactions to Accents. Educational Review 22, no. 3: 212-227.
Hartley, L. 2002. A View from the West: Perceptions of U.S. Dialects by Oregon Residents. In Handbook
of Perceptual Dialectology, ed. D. R. Preston, 315-332. Amsterdam: John Benjamins.
Labov, W. . 1972. Sociolinguistic Patterns. Philadelphia: University of Pennsylvania Press.
Niedzielski, N. A., and D. R. Preston. 2003. Folk Linguistics. New York: Walter de Gruyter.
Niedzielski, N. A.. 1999. The Effect of Social Information on the Perception of Sociolinguistic Variables .
Journal of Language and Social Psychology 18, no. 1: 62-85.
Preston, D. R., ed. 2002. Handbook of Perceptual Dialectology. Ed. D. R. Preston. Amsterdam: John
Benjamins Publishing Company.
Purnell, T., W. Idsardi, and J. Baugh. 1999. Perceptual and Phonetic Experiments on American English
Dialect Identification. Journal of Language and Social Psychology 18, no. 1: 10.
Thakerar, J. N., and H. Giles. 1981. They are - so they spoke. Language and Communcation 1, no. 2/3:
255-261.
Wardhaugh, R. 1999. Proper English: Myths and Misunderstandings about Language. Malden, MA:
Blackwell Publishers.
Wolfram, W. 1974. The Study of Social Dialects in American English. Englewood Cliffs, N.J: PrenticeHall.
.