Tokenise IPA transcriptions

In phonetisr: A Naive IPA Tokeniser

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Basic usage

library(phonetisr)

The main function of phonetisr is phonetise(). This function takes a character vector with IPA transcriptions and splits them into phones.

ipa <- c("pʰãkʰ", "tʰum̥", "ɛkʰɯ")

phonetise(ipa)

The default settings will tokenise each IPA symbol separately, rather than each phone, because by default phonetisr has no concept of "phone". IPA diacritics are then tokenised separately. The user can set diacritics = TRUE to automatically tokenise all diacritics with the preceding symbol (will not work of course for diacritics that are placed before the base symbol).

phonetise(ipa, diacritics = TRUE)

For the example words above, using diacritics = TRUE suffices. But what if you want more control? You can use the multi argument to specify phones that are made of multiple characters.

ph <- c("pʰ", "tʰ", "kʰ", "ã", "m̥")

phonetise(ipa, multi = ph)

In some cases you don't want a list of tokenised phones, but a vector where phones are separated by a specified character (like space, or a dot). You can set split = FALSE and the default separator (a space) will be used to separate phones in the resulting string. The separator character can be specified with sep.

phonetise(ipa, multi = ph, split = FALSE, sep = ".")

Using tibbles

A common use case of the phonetise() function is with tibble columns that have IPA transcriptions. The phonetisr package comes with kl_swades, a tibble with 195 Klingon words and their IPA transcription.

library(tidyverse)
data("kl_swadesh")
kl_swadesh

Let's phonetise the ipa column. We first want to define multi-character phones.

kl_multi <- c(
  "pʰ", "tʰ", "qʰ",
  "tɬ", "tʃ", "qχ", 
  "dʒ"
)

Then, we can use mutate() to create a new column phones with phonetised transcriptions.

kl_swadesh <- kl_swadesh |> 
  mutate(
    phones = phonetise(ipa, multi = kl_multi)
  )

kl_swadesh

phones is a "list" column, so kl_swadesh is "nested": see the Nested data vignette of tidyr.

A common operation is to count the number of occurrences of each phone. We can easily do that by first "unnesting" the phones column.

kl_unnest <- kl_swadesh |> 
  unnest(phones)
kl_unnest

Then we can count each phone with count().

kl_unnest |> 
  count(phones, sort = TRUE)

And also plot the counts.

kl_unnest |> 
  count(phones, sort = TRUE) |> 
  ggplot(aes(reorder(phones, desc(n)), n)) +
  geom_bar(stat = "identity")

Add phonetic features

The package comes with a function featurise() which takes a vector or list of phonetised words and returns a tibble with counts and phonetic features of the phones.

kl_feat <- featurise(kl_swadesh$phones)
kl_feat

We can use the info for plotting.

kl_feat |> 
  ggplot(aes(reorder(phone, desc(count)), count, fill = type)) +
  geom_bar(stat = "identity")

featurise() uses the info stored in ipa_symbols, which comes with the package. Note that the phonetic features included are only based on the IPA tables: the package is not aware of language-specific features.

data("ipa_symbols")
ipa_symbols

phonetisr documentation built on April 3, 2025, 10:49 p.m.