Another near-tie: 26 states say car-ml and 25 say carra-mel. The three-syllable pronunciation holds the east, the two-syllable one the west and the midwest, and the border runs almost straight down the middle.
with 2 syllables ("car-ml") leads 26 states
with 3 syllables ("carra-mel") leads 25 states
Each square is one state, placed roughly where it sits on the map and coloured by the answer that leads there. Survey question: “caramel”
Share of respondents in each state giving each answer. Percentages do not sum to 100: the survey offered answers this map does not name, and they are left out rather than folded into one of the three.
Caramel: Two Syllables or Three? — Harvard Dialect Survey.
State
with 2 syllables ("car-ml")
with 3 syllables ("carra-mel")
Alabama
20.8%
65.6% ◀
Alaska
64.6% ◀
16.7%
Arizona
49.2% ◀
22.1%
Arkansas
48.2% ◀
27%
California
34.8%
36.6% ◀
Colorado
54.4% ◀
18%
Connecticut
27.5%
53.5% ◀
Delaware
27.1%
37.5% ◀
Florida
25.7%
50.2% ◀
Georgia
20.5%
57.6% ◀
Hawaii
8.9%
75% ◀
Idaho
57.6% ◀
18%
Illinois
59.6% ◀
19%
Indiana
56.5% ◀
18%
Iowa
67.7% ◀
12.7%
Kansas
65.8% ◀
14.5%
Kentucky
33.3%
45.4% ◀
Louisiana
12.7%
69.4% ◀
Maine
24.3%
56.8% ◀
Maryland
17.3%
65.9% ◀
Massachusetts
22.3%
57% ◀
Michigan
65.2% ◀
14.2%
Minnesota
66.9% ◀
14.1%
Mississippi
18.4%
67.2% ◀
Missouri
64.2% ◀
15.4%
Montana
74.2% ◀
9.4%
Nebraska
61.3% ◀
12.7%
Nevada
55.2% ◀
22.4%
New Hampshire
10.7%
67.2% ◀
New Jersey
21.3%
56.8% ◀
New Mexico
52.1% ◀
26.9%
New York
23.9%
55.1% ◀
North Carolina
20.3%
59.4% ◀
North Dakota
69.8% ◀
12.6%
Ohio
57.3% ◀
20.2%
Oklahoma
51.5% ◀
24.9%
Oregon
54.3% ◀
21.1%
Pennsylvania
47% ◀
27.9%
Rhode Island
20.2%
51.9% ◀
South Carolina
24%
54.8% ◀
South Dakota
73.5% ◀
10.9%
Tennessee
22.4%
59.8% ◀
Texas
36.7%
41.4% ◀
Utah
57.6% ◀
14.5%
Vermont
26.6%
44.3% ◀
Virginia
21.4%
59% ◀
Washington
58% ◀
17.5%
Washington, D.C.
20.9%
58.1% ◀
West Virginia
34.7%
39.4% ◀
Wisconsin
69.9% ◀
11.5%
Wyoming
66.2% ◀
15.5%
Where this data comes from
The map above is drawn from the Harvard Dialect Survey, collected by Bert Vaux and Scott Golder in 2002–2003. It asked 122 questions and recorded, for each one, the share of respondents in every US state who gave each answer. It is a large self-selected sample rather than a census, and it is now two decades old — treat it as a well-supported picture of how people spoke then, not a measurement of today.
Data used under the CC BY-NC-SA 3.0 licence: attribution and share-alike, non-commercial. DialectGuessr’s own survey data is separate and is described on the about page.
