AMR/R/translate.R

218 lines
10 KiB
R
Executable File

# ==================================================================== #
# TITLE #
# Antimicrobial Resistance (AMR) Data Analysis for R #
# #
# SOURCE #
# https://github.com/msberends/AMR #
# #
# LICENCE #
# (c) 2018-2022 Berends MS, Luz CF et al. #
# Developed at the University of Groningen, the Netherlands, in #
# collaboration with non-profit organisations Certe Medical #
# Diagnostics & Advice, and University Medical Center Groningen. #
# #
# This R package is free software; you can freely use and distribute #
# it for both personal and commercial purposes under the terms of the #
# GNU General Public License version 2.0 (GNU GPL-2), as published by #
# the Free Software Foundation. #
# We created this package for both routine data analysis and academic #
# research and it was publicly released in the hope that it will be #
# useful, but it comes WITHOUT ANY WARRANTY OR LIABILITY. #
# #
# Visit our website for the full manual and a complete tutorial about #
# how to conduct AMR data analysis: https://msberends.github.io/AMR/ #
# ==================================================================== #
#' Translate Strings from AMR Package
#'
#' For language-dependent output of AMR functions, like [mo_name()], [mo_gramstain()], [mo_type()] and [ab_name()].
#' @inheritSection lifecycle Stable Lifecycle
#' @details Strings will be translated to foreign languages if they are defined in a local translation file. Additions to this file can be suggested at our repository. The file can be found here: <https://github.com/msberends/AMR/blob/main/data-raw/translations.tsv>. This file will be read by all functions where a translated output can be desired, like all [`mo_*`][mo_property()] functions (such as [mo_name()], [mo_gramstain()], [mo_type()], etc.) and [`ab_*`][ab_property()] functions (such as [ab_name()], [ab_group()], etc.).
#'
#' Currently supported languages are: `r vector_and(names(LANGUAGES_SUPPORTED), quotes = FALSE)`. All these languages have translations available for all antimicrobial agents and colloquial microorganism names.
#'
#' Please suggest your own translations [by creating a new issue on our repository](https://github.com/msberends/AMR/issues/new?title=Translations).
#'
#' ## Changing the Default Language
#' The system language will be used at default (as returned by `Sys.getenv("LANG")` or, if `LANG` is not set, [Sys.getlocale()]), if that language is supported. But the language to be used can be overwritten in two ways and will be checked in this order:
#'
#' 1. Setting the R option `AMR_locale`, e.g. by running `options(AMR_locale = "de")`
#' 2. Setting the system variable `LANGUAGE` or `LANG`, e.g. by adding `LANGUAGE="de_DE.utf8"` to your `.Renviron` file in your home directory
#'
#' Thus, if the R option `AMR_locale` is set, the system variables `LANGUAGE` and `LANG` will be ignored.
#' @inheritSection AMR Read more on Our Website!
#' @rdname translate
#' @name translate
#' @export
#' @examples
#' # The 'language' argument of below functions
#' # will be set automatically to your system language
#' # with get_AMR_locale()
#'
#' # English
#' mo_name("CoNS", language = "en")
#' #> "Coagulase-negative Staphylococcus (CoNS)"
#'
#' # Danish
#' mo_name("CoNS", language = "da")
#' #> "Koagulase-negative stafylokokker (KNS)"
#'
#' # Dutch
#' mo_name("CoNS", language = "nl")
#' #> "Coagulase-negatieve Staphylococcus (CNS)"
#'
#' # German
#' mo_name("CoNS", language = "de")
#' #> "Koagulase-negative Staphylococcus (KNS)"
#'
#' # Italian
#' mo_name("CoNS", language = "it")
#' #> "Staphylococcus negativo coagulasi (CoNS)"
#'
#' # Portuguese
#' mo_name("CoNS", language = "pt")
#' #> "Staphylococcus coagulase negativo (CoNS)"
#'
#' # Spanish
#' mo_name("CoNS", language = "es")
#' #> "Staphylococcus coagulasa negativo (SCN)"
get_AMR_locale <- function() {
# AMR versions 1.3.0 and prior used the environmental variable:
if (!identical("", Sys.getenv("AMR_locale"))) {
options(AMR_locale = Sys.getenv("AMR_locale"))
}
if (!is.null(getOption("AMR_locale", default = NULL))) {
lang <- getOption("AMR_locale")
if (lang %in% LANGUAGES_SUPPORTED) {
return(lang)
} else {
stop_("unsupported language set as option 'AMR_locale': \"", lang, "\" - use either ",
vector_or(paste0('"', LANGUAGES_SUPPORTED, '" (', names(LANGUAGES_SUPPORTED), ")"), quotes = FALSE))
}
} else {
# now check the LANGUAGE system variable - return it if set
if (!identical("", Sys.getenv("LANGUAGE"))) {
return(coerce_language_setting(Sys.getenv("LANGUAGE")))
}
if (!identical("", Sys.getenv("LANG"))) {
return(coerce_language_setting(Sys.getenv("LANG")))
}
}
# fallback - automatic determination based on LC_COLLATE
if (interactive() && message_not_thrown_before("get_AMR_locale", entire_session = TRUE)) {
lang <- coerce_language_setting(Sys.getlocale("LC_COLLATE"))
if (lang != "en") {
message_("Assuming the ", names(LANGUAGES_SUPPORTED)[LANGUAGES_SUPPORTED == lang],
" language for the AMR package. Change this with `options(AMR_locale = \"...\")` or see `?get_AMR_locale()`. ",
"Supported languages are ", vector_and(names(LANGUAGES_SUPPORTED), quotes = FALSE),
". This note will be shown once per session.")
}
return(lang)
}
coerce_language_setting(Sys.getlocale("LC_COLLATE"))
}
coerce_language_setting <- function(lang) {
# grepl() with ignore.case = FALSE is 8x faster than %like_case%
if (grepl("^(English|en_|EN_)", lang, ignore.case = FALSE, perl = TRUE)) {
# as first option to optimise speed
"en"
} else if (grepl("^(German|Deutsch|de_|DE_)", lang, ignore.case = FALSE, perl = TRUE)) {
"de"
} else if (grepl("^(Dutch|Nederlands|nl_|NL_)", lang, ignore.case = FALSE, perl = TRUE)) {
"nl"
} else if (grepl("^(Danish|Dansk|da_|DA_)", lang, ignore.case = FALSE, perl = TRUE)) {
"da"
} else if (grepl("^(Spanish|Espa.+ol|es_|ES_)", lang, ignore.case = FALSE, perl = TRUE)) {
"es"
} else if (grepl("^(Italian|Italiano|it_|IT_)", lang, ignore.case = FALSE, perl = TRUE)) {
"it"
} else if (grepl("^(French|Fran.+ais|fr_|FR_)", lang, ignore.case = FALSE, perl = TRUE)) {
"fr"
} else if (grepl("^(Portuguese|Portugu.+s|pt_|PT_)", lang, ignore.case = FALSE, perl = TRUE)) {
"pt"
} else if (grepl("^(Russian|pycc|ru_|RU_)", lang, ignore.case = FALSE, perl = TRUE)) {
"ru"
} else if (grepl("^(Swedish|Svenskt|sv_|SV_)", lang, ignore.case = FALSE, perl = TRUE)) {
"sv"
} else {
# other language -> set to English
"en"
}
}
# translate strings based on inst/translations.tsv
translate_AMR <- function(from,
language = get_AMR_locale(),
only_unknown = FALSE,
only_affect_ab_names = FALSE,
only_affect_mo_names = FALSE) {
if (is.null(language)) {
return(from)
}
if (language %in% c("en", "", NA)) {
return(from)
}
df_trans <- TRANSLATIONS # internal data file
from.bak <- from
from_unique <- unique(from)
from_unique_translated <- from_unique
stop_ifnot(language %in% LANGUAGES_SUPPORTED,
"unsupported language: \"", language, "\" - use either ",
vector_or(LANGUAGES_SUPPORTED, quotes = TRUE),
call = FALSE)
# only keep lines where translation is available for this language
df_trans <- df_trans[which(!is.na(df_trans[, language, drop = TRUE])), , drop = FALSE]
# and where the original string is not equal to the string in the target language
df_trans <- df_trans[which(df_trans[, "pattern", drop = TRUE] != df_trans[, language, drop = TRUE]), , drop = FALSE]
if (only_unknown == TRUE) {
df_trans <- subset(df_trans, pattern %like% "unknown")
}
if (only_affect_ab_names == TRUE) {
df_trans <- subset(df_trans, affect_ab_name == TRUE)
}
if (only_affect_mo_names == TRUE) {
df_trans <- subset(df_trans, affect_mo_name == TRUE)
}
if (NROW(df_trans) == 0) {
return(from)
}
# default: case sensitive if value if 'case_sensitive' is missing:
df_trans$case_sensitive[is.na(df_trans$case_sensitive)] <- TRUE
# default: not using regular expressions if 'regular_expr' is missing:
df_trans$regular_expr[is.na(df_trans$regular_expr)] <- FALSE
# check if text to look for is in one of the patterns
any_form_in_patterns <- tryCatch(
any(from_unique %like% paste0("(", paste(gsub(" +\\(.*", "", df_trans$pattern), collapse = "|"), ")")),
error = function(e) {
warning_("Translation not possible. Please open an issue on GitHub (https://github.com/msberends/AMR/issues).")
return(FALSE)
})
if (NROW(df_trans) == 0 | !any_form_in_patterns) {
return(from)
}
lapply(seq_len(nrow(df_trans)),
function(i) from_unique_translated <<- gsub(pattern = df_trans$pattern[i],
replacement = df_trans[i, language, drop = TRUE],
x = from_unique_translated,
ignore.case = !df_trans$case_sensitive[i] & df_trans$regular_expr[i],
fixed = !df_trans$regular_expr[i],
perl = df_trans$regular_expr[i]))
# force UTF-8 for diacritics
from_unique_translated <- enc2utf8(from_unique_translated)
# a kind of left join to get all results back
from_unique_translated[match(from.bak, from_unique)]
}