By Kevin Ryan
Last updated (release 1.01) June 30, 2020.
Please feel free to email me with errors/bugs/improvements.
Some dictionaries provide word pronunciations in IPA
Phonetic Alphabet). This app improves on those transcriptions by offering:
- A form of IPA consistent with university linguistics courses
- Narrow transcriptions (e.g. minor diphthongs) and allophonic detail (e.g. aspiration) that
dictionaries usually omit
- The option to toggle various allophonic processes
- The ability to highlight selected features/natural classes
- More pronunciation variants than the typical dictionary provides
- Inflected forms of words not normally given separate entries in dictionaries
- Batch processing of words
Transcriptions here are ultimately based on the
CMU Pronouncing Dictionary,
but modified to render ARPABET in IPA, including
modifications to bring the transcription into line with standard phonetics textbooks such as
Ladefoged & Johnson, A course in phonetics (7th ed., 2015), and to add allophonic detail,
consistent with a narrower level of transcription. The dialect is roughly General American. The feature
system follows Bruce Hayes, Introductory phonology (2009). Features are generally binary,
classifying every phone as + or -. On this system, features can also have a value of 0 (not
applicable). For example, [anterior] is defined only for coronals; for all non-coronals, it is 0.
Neither + nor - matches 0.
Variant pronunciations of words can be accessed by clicking on the superscript gray numbers in the IPA
rendering. Hover over (or tap) the names of allophonic processes for details about how each rule is
- Hayes (2009) is not the only proposed feature system; others' classifications vary on some points
(but usually not much). Additionally, diphthongs and affricates are treated as units here for feature
assignment, which in some cases requires resolving conflicts between the two internal symbols (e.g. [ɪ]
by itself is lax, but highlighted as tense when part of [eɪ]).
- The app does not do any morphological parsing. In reality, the pronunciation of some words depends on
their syntactic category. For example, permit is initially stressed as a noun, but finally as a
verb. This app offers both variants, regardless of context.
- Each word is given in its citation form. In real speech, words are often reduced. CMU captures some of
these possibilities in its variants, but certainly not all (or even most). This app does not take into
account prosody or sandhi (sentence-level phonology and juncture). Rather, it treats each word as if it
were pronounced in isolation. This is done to facilitate checking word lists, but is
not realistic of running speech.
- Finally, since the app does only minimal morphological parsing (basically, s-stripping), it misses many
items, including those that are transparently related to forms it does recognize. For instance,
diversities is not found, but diversity is. It would be straightforward to derive the
pronunciation of the former from the latter, but the app does not attempt it. In general, one can
search for similar or related words if the desired word is not available. Fortunately, CMU already
includes a huge number of inflected items.