In bilingual-project/jtracer: An R interface to jTRACE

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

library(jtracer)
jtrace_install()

At the lowest level of the TRACE model's architecture lie the acoustic features associated with the phonemes that compose words in the lexicon. jTRACE considers seven feature dimensions (see ?phonemes): power (pow), vocalic (vow), diffusiveness (dif), acuteness (acu), consonantal (con), voicing (voi), and burst (bur). These are the original dimensions used by McClelland and Elman (1986) in their original implementation of TRACE. Prior to the simulation, the model is introduced to a set of familiar phonemes, mimicking its familiarity with a language. Each of these phonemes are coded by assigning them a 1-9 score in each of the feature dimensions. McClelland and Elman's original phoneme set was later extended by Mayor and Plunkett (2014).

library(jtracer)
data("phonemes")
Encoding(phonemes[["ipa"]]) <- "Latin-1"
Encoding(phonemes[["trace"]]) <- "Latin-1"

head(phonemes)

Non-English phonemes

Both versions were explicitly designed to fit the English phonology. We have coded new phonemes, not restricted to the English phonology, in the phonemes data frame included in this package. This way, it is possible to run simlation on a model that is familiar to phonologies different than the English one, such as Spanish and Catalan. The dataframe includes three logical variables indicating whether a given phoneme is present in English (is_english), Spanish (is_spanish), or Catalan (is_catalan), respectively.

English phoneme set

library(gt)
library(dplyr)

phonemes %>%
  filter(is_english) %>% 
  select(-starts_with("is_")) %>% 
  rename_at(vars(pow:bur), toupper) %>% 
  gt(rowname_col = "id", groupname_col = "type") %>% 
  tab_spanner(md("**Acoustic features**"), columns = 6:12) %>% 
  cols_label(
    ipa = md("**IPA symbol**"),
    trace = md("**jTRACE symbol**"),
    description = ""
  ) %>% 
  data_color(
    columns = 6:12,
    colors = scales::col_numeric(
      palette = c(
        "white", "orange"), 
      domain = c(0, 9)
    )
  )

Spanish phoneme set

phonemes %>%
  filter(is_spanish) %>% 
  select(-starts_with("is_")) %>% 
  rename_at(vars(pow:bur), toupper) %>% 
  gt(rowname_col = "id", groupname_col = "type") %>% 
  tab_spanner(md("**Acoustic features**"), columns = 6:12) %>% 
  cols_label(
    ipa = md("**IPA symbol**"),
    trace = md("**jTRACE symbol**"),
    description = ""
  ) %>% 
  data_color(
    columns = 6:12,
    colors = scales::col_numeric(
      palette = c(
        "white", "orange"), 
      domain = c(0, 9)
    )
  )

Catalan phoneme set

phonemes %>%
  filter(is_catalan) %>% 
  select(-starts_with("is_")) %>% 
  rename_at(vars(pow:bur), toupper) %>% 
  gt(rowname_col = "id", groupname_col = "type") %>% 
  tab_spanner(md("**Acoustic features**"), columns = 6:12) %>% 
  cols_label(
    ipa = md("**IPA symbol**"),
    trace = md("**jTRACE symbol**"),
    description = ""
  ) %>% 
  data_color(
    columns = 6:12,
    colors = scales::col_numeric(
      palette = c(
        "white", "orange"), 
      domain = c(0, 9)
    )
  )

Creating a jTRACE-compatible language from the `phonemes` dataset

Imagine we wanted to run some simulations on a model that is familiar to the Spanish phonology. We could do so by shortlisting those phonemes present in Spanish (is_spanish is TRUE) and then following the steps in the Creating a language vignette. Please note that allophonic relations and duration scalars cannot be yet modified.

spanish_phonemes <- phonemes %>% 
  filter(is_spanish) %>% 
  select(trace, pow:bur)

jtrace_create_language(
  language_name = "spanish",
  features = spanish_phonemes
)

Now we are ready to launch jTRACE running jtrace_launch() and loading the languages/spanish.xml file in the Phonemes tab.

How are the seven feature dimensions related to each other?

library(ggplot2)
library(ggcorrplot)

corr <- cor_pmat(phonemes[, 9:15])
colnames(corr) <- toupper(colnames(corr))
rownames(corr) <- toupper(colnames(corr))

ggcorrplot(
  corr, type = "upper", hc.order = TRUE, outline.color = "grey",
  lab = TRUE, colors = c("#f76560", "grey90", "#60f777"), show.diag = TRUE
) +
  theme_minimal() +
  theme(
    legend.position = "right",
    axis.title = element_blank(),
    axis.text = element_text(colour = "black"),
    panel.grid = element_blank()
  )

Appendix

This is the complete list of implemented phonemes:

phonemes %>% 
  rename_at(vars(pow:bur), toupper) %>%
  rename_at(vars(starts_with("is_")), ~gsub("is_", "", .)) %>% 
  gt(rowname_col = "id", groupname_col = "type") %>% 
  tab_spanner(md("**Acoustic features**"), columns = 6:12) %>% 
  cols_label(
    ipa = md("**IPA**"),
    trace = md("**jTRACE**"),
    description = ""
  ) %>% 
  data_color(
    columns = 6:12,
    colors = scales::col_numeric(
      palette = c(
        "white", "orange"), 
      domain = c(0, 9)
    )
  )