phonfeatures is an R package which allows the user to efficiently add
columns with articulatory features to existing data frames with phonetic
characters using the function add_features().
The package provides several generic features for all regular IPA and
X-SAMPA characters as well as for most diacritics. These generic
features can be checked using the function feature_lookup(). The
package also provides functionality for changing the feature values of
known characters with feature_reassign(), and for introducing unknown
characters with feature_assign().
You can install the development version of phonfeatures from
GitHub with:
# install.packages("devtools")
devtools::install_github("rpuggaardrode/phonfeatures")The most common use case is when you have a data frame containing transcriptions, and you want to convert those transcriptions into some articulatory features. It could fx look like this:
library(phonfeatures)
head(test_data, 10)
#> Identifier IntVocIntervals tmin tmax age Stop
#> 1 1 1 0.000000 0.245914 20 d
#> 2 2 2 0.245914 0.798165 20 k
#> 3 3 3 0.798165 1.103336 20 d
#> 4 4 4 1.103336 1.716694 20 k
#> 5 5 5 1.716694 2.206374 20 k
#> 6 6 6 2.206374 2.546418 20 g
#> 7 7 7 2.546418 2.844114 20 t
#> 8 8 8 2.844114 3.357136 20 k
#> 9 9 9 3.357136 3.962371 20 k
#> 10 10 10 3.962371 4.462383 20 kSay you want to add columns to this data frame with information about place of articulation and laryngeal features corresponding to the stop column. You would proceed like this:
new_data <- add_features(data=test_data, col='Stop', feature=c('place', 'lar'))
head(new_data, 10)
#> Identifier IntVocIntervals tmin tmax age Stop place lar
#> 1 1 1 0.000000 0.245914 20 d alveolar voiced
#> 2 2 2 0.245914 0.798165 20 k velar voiceless
#> 3 3 3 0.798165 1.103336 20 d alveolar voiced
#> 4 4 4 1.103336 1.716694 20 k velar voiceless
#> 5 5 5 1.716694 2.206374 20 k velar voiceless
#> 6 6 6 2.206374 2.546418 20 g velar voiced
#> 7 7 7 2.546418 2.844114 20 t alveolar voiceless
#> 8 8 8 2.844114 3.357136 20 k velar voiceless
#> 9 9 9 3.357136 3.962371 20 k velar voiceless
#> 10 10 10 3.962371 4.462383 20 k velar voicelessYou can use ?add_features to see which features are generically
available, or use feature_lookup() to check all the features that are
generically assigned to a character:
feature_lookup(phon='g')
#> height backness roundness place major_place manner major_manner lar voice
#> 8 velar dorsal stop obstruent voiced voiced
#> length modifications syllabic release nasalization tone
#> 8 short NA no released no NAfeature_lookup() can also be used to check specific feature values:
feature_lookup(phon='g', feature=c('lar', 'place'))
#> lar place
#> 8 voiced velarThe generic features will not be suitable for all purposes. For example, if (like me) you are interested in Danish stops, there’s a good chance that you have been using /b d g p t k/ as a shorthand for sounds that are not actually voiced and voiceless, but rather voiceless and aspirated.
The package also provides functions for creating a new lookup table used
for checking feature values. We can update the value of lar for /b d
g/ with the function feature_reassign() and save it to a new data
frame dan_lkup like this:
dan_lkup <- feature_reassign(sampa=c('b', 'd', 'g'),
feature='lar', val='voiceless')And the values of /p t k/ can then be updated by running
feature_reassign() again, this time specifying that we’re using the
lookup table dan_lkup:
dan_lkup <- feature_reassign(sampa=c('p', 't', 'k'),
feature='lar', val='aspirated',
lookup=dan_lkup)We can now use dan_lkup to add features to our data frame, instead of
using the generic features:
new_data <- add_features(data=test_data, col='Stop', feature=c('place', 'lar'),
lookup=dan_lkup)
head(new_data, 10)
#> Identifier IntVocIntervals tmin tmax age Stop place lar
#> 1 1 1 0.000000 0.245914 20 d alveolar voiceless
#> 2 2 2 0.245914 0.798165 20 k velar aspirated
#> 3 3 3 0.798165 1.103336 20 d alveolar voiceless
#> 4 4 4 1.103336 1.716694 20 k velar aspirated
#> 5 5 5 1.716694 2.206374 20 k velar aspirated
#> 6 6 6 2.206374 2.546418 20 g velar voiceless
#> 7 7 7 2.546418 2.844114 20 t alveolar aspirated
#> 8 8 8 2.844114 3.357136 20 k velar aspirated
#> 9 9 9 3.357136 3.962371 20 k velar aspirated
#> 10 10 10 3.962371 4.462383 20 k velar aspiratedYou may have noticed that the phonetic character argument in
feature_reassign() is called sampa. This is because this function
only works with X-SAMPA
characters. add_features() and feature_lookup() also use X-SAMPA
characters by default, but these functions have an option to use IPA
instead, by setting ipa=TRUE. Let’s see what this does to the code we
just ran:
new_data <- add_features(data=test_data, col='Stop', feature=c('place', 'lar'),
lookup=dan_lkup, ipa=TRUE)
head(new_data, 10)
#> Identifier IntVocIntervals tmin tmax age Stop place lar
#> 1 1 1 0.000000 0.245914 20 d alveolar voiceless
#> 2 2 2 0.245914 0.798165 20 k velar aspirated
#> 3 3 3 0.798165 1.103336 20 d alveolar voiceless
#> 4 4 4 1.103336 1.716694 20 k velar aspirated
#> 5 5 5 1.716694 2.206374 20 k velar aspirated
#> 6 6 6 2.206374 2.546418 20 g velar voiceless
#> 7 7 7 2.546418 2.844114 20 t alveolar aspirated
#> 8 8 8 2.844114 3.357136 20 k velar aspirated
#> 9 9 9 3.357136 3.962371 20 k velar aspirated
#> 10 10 10 3.962371 4.462383 20 k velar aspiratedIt makes no difference of course! /b d g p t k/ are identical in IPA and
X-SAMPA. But this is what happens when feature_lookup() is used with
an IPA character which doesn’t exist as an X-SAMPA character.
feature_lookup(phon='ø', feature='length')
#> Error in feature_lookup(phon = "ø", feature = "length"): I am unfamiliar with ø
#> Definitions for unknown characters can be added with the feature_assign() function.The function throws a warning saying that the character is unknown.
Let’s try again with ipa=TRUE:
feature_lookup(phon='ø', feature='length', ipa=TRUE)
#> [1] "short"Say you have a data frame with IPA characters, and the sound /ø/ is
always long in the language, so you want to change the length value to
long. feature_reassign() only accepts X-SAMPA characters, but fear
not! The function ipa::ipa() is used under the hood to convert IPA
characters to X-SAMPA characters, so you can simply reassign the
corresponding X-SAMPA character (in this case 2):
lkup <- feature_reassign(sampa='2', feature='length', val='long')
feature_lookup(phon='ø', feature='length', lookup=lkup, ipa=TRUE)
#> [1] "long"You may want to use add_features() with a data frame where, for one
reason or another, you have been using non-standard symbols. For
example, you’ve been using ph instead of the standard X-SAMPA p_h to
indicate an aspirated bilabial stop. In this case, feature_lookup()
will shoot you an error:
feature_lookup(phon='ph')
#> Error in feature_lookup(phon = "ph"): I am unfamiliar with ph
#> Definitions for unknown characters can be added with the feature_assign() function.Again, fear not! The feature_assign function can help with creating a
customized lookup table with new characters. This feature also allows
you to copy the features from a known character, or a character with
known diacritics, using the copy argument. For our ph case, you
would do the following:
lkup <- feature_assign(new='ph', copy='p_h')
feature_lookup(phon='ph', lookup=lkup)
#> height backness roundness place major_place manner major_manner
#> 103 bilabial labial stop obstruent
#> lar voice length modifications syllabic release nasalization
#> 103 aspirated voiceless short NA no released no
#> tone
#> 103 NAfeature_assign also allows you to build your own feature sets from
scratch:
lkup <- feature_assign(new='kh',
feature=c('place', 'manner', 'lar'),
val=c('velar', 'stop', 'aspirated'))
feature_lookup(phon='kh', lookup=lkup)
#> height backness roundness place major_place manner major_manner lar
#> 103 <NA> <NA> <NA> velar <NA> stop <NA> aspirated
#> voice length modifications syllabic release nasalization tone
#> 103 <NA> <NA> NA <NA> <NA> <NA> NAOr you can copy a known character and change some of the features (say,
if you’ve been using th to refer to a long, dental aspirated stop):
lkup <- feature_assign(new='th',
feature=c('place', 'length'),
val=c('dental', 'long'),
copy='t_h')
feature_lookup(phon='th', lookup=lkup)
#> height backness roundness place major_place manner major_manner lar
#> 103 dental coronal stop obstruent aspirated
#> voice length modifications syllabic release nasalization tone
#> 103 voiceless long NA no released no NAYou can bulk add new characters using feature_assign assuming they can
all be copied from known characters, or that the same feature needs to
be manipulated for all of them. In the following code block, features
are bulk added from known characters:
lkup <- feature_assign(new=c('ph', 'th', 'kh'),
copy=c('p_h', 't_h', 'k_h'))
tail(lkup, 3)
#> segm height backness roundness place major_place manner major_manner
#> 103 ph bilabial labial stop obstruent
#> 104 th alveolar coronal stop obstruent
#> 105 kh velar dorsal stop obstruent
#> lar voice length modifications syllabic release nasalization
#> 103 aspirated voiceless short NA no released no
#> 104 aspirated voiceless short NA no released no
#> 105 aspirated voiceless short NA no released no
#> tone
#> 103 NA
#> 104 NA
#> 105 NASay you have been using ph, th, kh for preaspirated stop. They can
be bulk added like this:
lkup <- feature_assign(new=c('hp', 'ht', 'hk'),
feature='lar',
val='preaspirated',
copy=c('p', 't', 'k'))
tail(lkup, 3)
#> segm height backness roundness place major_place manner major_manner
#> 103 hp bilabial labial stop obstruent
#> 104 ht alveolar coronal stop obstruent
#> 105 hk velar dorsal stop obstruent
#> lar voice length modifications syllabic release nasalization
#> 103 preaspirated voiceless short NA no released no
#> 104 preaspirated voiceless short NA no released no
#> 105 preaspirated voiceless short NA no released no
#> tone
#> 103 NA
#> 104 NA
#> 105 NAUnfortunately, feature_assign does not at present allow you to add
unknown IPA characters.
If you run into any problems using phonfeatures, feel free to leave a
bug report on GitHub or
reach out at r.puggaard at phonetik.uni-muenchen.de