The IPA Translation App: a wonderful impossibility

This was originally posted on my blog, Notes from a Linguistic Mystic in 2007. See all posts

Note that more often than ever, people are making online ‘lookup’ tools for IPA, which have vast lists of words with their (often poorly transcribed) IPA equivalents in a mishmash of dialects. This remains true! - February 2023!

So, I’m somewhat obsessed with checking the statistics of who comes here, who gets referred from where, and what search terms they used to find me. Well, the other day, somebody came here from google searching for “IPA translation widget”. For those of you unfamiliar with the terms, a “widget” is a small computer program, and IPA refers to the International Phonetic Alphabet. What this person seems to be wanting was a widget that, like some existing translation widgets, could take a block of text and immediately turn it into IPA characters. For the first few moments, I thought “Wow! That’d be a great idea!”.

Now, as somebody who uses the IPA very, very frequently, such a thing would be wonderful if it worked well. However, I think it would be impossible to actually create a program that goes from English writing to IPA transcriptions without incredible advances in Artificial Intelligence and speech recognition. Here’s why…

Transcription, not translation

At the surface, this doesn’t seem so crazy. Apple includes a widget to do rough, automated translations with Dashboard, and although I never trust automated translations, it does alright for basic words and phrases. I suspect that our anonymous searcher saw that widget and thought “Wow, cool! I wonder if it can help me put something into the IPA”. However, the fundamental difference between translating a sentence into Spanish and putting that same sentence into the IPA is that the IPA isn’t really a language at all, but instead, it’s a method of writing sounds.

The International Phonetic Alphabet is really a set of symbols, each of which represents a sound, sound characteristic, or other element of spoken language. What the IPA allows a linguist (or speech pathologist, or teacher…) to do is to take spoken language and put it onto paper (‘transcription’) with a great deal more precision than most other writing systems. The IPA isn’t a language in itself, it’s just an alternative, phonetic writing system for other languages. The beauty of this is that the IPA is designed to be able to be used not just for English, but for any language. The IPA symbols can be used to transcribe sounds not just from English, but from languages all over the world.

Broad vs. Narrow Transcription

The IPA can be used to transcribe sounds with two different degrees of precision.

If one takes advantage of all the symbols and diacritics, one can make a “narrow” or “phonetic” transcription. At this level, the linguist aims to capture all the detail possible about the word or phrase, including variations across word boundaries, sounds that occur in speech but are unnoticed or unrecognized by native speakers, and even features like intonation and pauses. From these transcriptions, a well-trained linguist could pronounce the words and phrases almost exactly as the speaker did, based simply on the transcriptions. The first, smallest line in the title graphic is a narrow transcription of me pronouncing the site’s title.

This degree of precision would be impossible for a modern computer widget to produce, simply because narrow transcriptions are based on actual words and phrases by a speaker, and really, one needs a fairly trained ear to make an accurate narrow transcription of a word or phrase. Sure, it could use a database of narrowly transcribed words from other speakers, but really, that’s not a narrow transcription. It’s not going to pick up on the variations that each speaker produces, like accents, vowel changes, unusual sound choices, or even tiny speech errors.

The alternative is called “broad” or “phonemic” transcription, expresses the basic sounds of a language or phrase, often more precisely than the native writing system, but at the same time, leaves out detail that’s not necessary to a native speaker. The middle line in the title graphic for this page is a phonemic transcription. Some dictionaries, including the built in OS X dictionary (if you enable IPA in Dictionary Preferences), can show you the standard american IPA Broad transcription form of a word.

Now, using a dictionary of words in a given language and their IPA equivalents, a computer could likely match things and give a passable broad transcription. However, there are variations that occur between people that show up even at a broad level, and are large enough to identify a speaker’s accent, dialect, or even idiolect. For some people (myself included), “caught” and “cot” have the same vowel, but for others, they’re two distinct vowels. So, even at a broad level, you’re not going to get any sort of reliable transcription of one’s actual speech from a computer widget, just a rough approximation.

Why are you transcribing anyways?

In the end, whether such a widget would be useful at all boils down to your reason for needing a transcription. Some people might be learning English and would want a better method of knowing how a given word is supposed to sound. For that, any good dictionary’s pronunciation key should do the trick.

Some people might be interested in the IPA, or want to know how a given word sounds. For that, they’d be better off getting a good phonetics textbook and learning a bit of the IPA themselves, along with some knowledge of phonetics.

However, our widget searcher might just be stuck in an introductory Linguistics course, having to transcribe their speech for an assignment. If so, I offer just one piece of advice: Don’t plagarize transcriptions off the web or from a dictionary. Your professor should have no trouble noticing if you’re not transcribing your own dialect, and everybody’s got a dialect.

Remember, if there’s one thing that phonetics professors are good at, it’s picking out a phone-y.