knitr::opts_chunk$set( collapse = TRUE, comment = "#>", fig.path = "README-" ) devtools::load_all()
autoscore
R Package:
r packageVersion("autoscore")
Shiny App: temporary location at https://tysonstanley.shinyapps.io/autoscore/
Authors:
The purpose of autoscore
is to automatically score word identification in speech perception research, such as studies involving listener understanding of speech in background noise or disordered speech. The program uses a flexible number of rules that determine whether a response set of words (i.e., listener transcriptions) match a target set of words (i.e., speech corpus). At the most basic level, Autoscore counts words in the listener transcript as correct if they match the words in the target phrase exactly (regardless of word order), or match a homophone or common misspelling of the target word. Individual rules can be applied or removed, depending on the needs of researcher and the scoring rules of the research lab. Examples of rules available in Autoscore include the ability to count as correct substitutions of articles (A for The) or differences in plural or tense (adding -s or -ed to a word). Additional rules can be added by the researcher as needed.
The rule options are categorized into either spelling rules or grammar rules.
acceptable_spell_rule
: Response word counted correct if it is a homophone or common misspelling of the target word, according to a preloaded default acceptable spelling list (contains over 300 common acceptable spellings). User can also download the default acceptable spelling list, add/remove items, and upload for automation. Response word counted correct if is on the acceptable spelling list. Default is FALSE
.root_word_rule
: Response word counted correct if the target word (e.g. ‘day’) is embedded at either the beginning (e.g. ‘daybreak’) of the target word. Default is FALSE
.double_letter_rule
: Response word counted correct if it omitted a double letter within a word (e.g. ‘atack’ matches ‘attack’) or added an unnecessary double letter (e.g. ‘occassion’ matches ‘occasion’). Default is FALSE
.tense_rule
: Response word counted correct if it differs from the target word only by tense; Default is FALSE
.tense_add_rule
: Response word counted correct if it only adds a tense as compared to the target word; Default is FALSE
. It is also referred to as "Tense+ Rule."plural_rule
: Response word counted correct if it differs from the target word only by plurality; Default is FALSE
.plural_add_rule
: Response word counted correct if it only adds a plural as compared to the target word; Default is FALSE
. It is also referred to as "Plural+ Rule."a_the_rule
: Substitutions between “a” and “the” to be scored as correct; Default is FALSE
.The API of the R
package is simple. A single call to autoscore()
with the formatted data will run everything for you. This function is a composite of several sub-functions that do various jobs:
select_cols()
-- The first function which takes the data and gets it in the right format for analysis.split_clean()
-- Using the cleaned data from select_cols()
, this uses stringr
to turn the phrases into individual words.alternate_fun()
-- If a data.frame of alternate spellings is provided, this function will find and normalize all alternate spellings to match the original spelling as defined by the researcher.match_position_basic()
-- This function is the workhorse of the package. It takes the cleaned data and does three main things: 1) applies all the rules except for the position_rule
, 2) finds the matches between the responses and the targets, and 3) reports how far away the matches are from each other.count_matches()
-- Finally, this function takes the information from match_position_basic()
and counts the number of matches based on the position_rule
.Beyond the main analysis when using autoscore()
, we can also call pwc()
to get the percent words correct (based on the number of target words) for each observation.
Visit https://tysonstanley.shinyapps.io/autoscore/ to use the online tool. Instructions for its use are found there.
To install the package use the developmental version as it is not yet on CRAN.
devtools::install_github("tysonstanley/autoscore_package")
An example of the use of autoscore
is below. We will use the example data set provided in the package.
library(tidyverse) library(autoscore) data("example_data") example_data
First, let's use all the defaults and look at the first 10 rows of the output.
example_data %>% autoscore() %>% ## using all the defaults as.tibble() ## to shorted output
Next, let's change some of the rules.
example_data %>% autoscore(plural_rule = FALSE, tense_rule = FALSE) %>% as.tibble()
We can also change the output type to "none" to get all the data from the computation.
example_data %>% autoscore(output = "none")
To use the acceptable spelling rule, let's use the default provided in autoscore
. . In the data frame below, the target
spellings are the generally accepted spellings that are in the target list of words while the acceptable
column are those that should also be counted as correct.
autoscore::acceptable_spellings
Using this, we can provide it to the autoscore()
function with the acceptable_df
argument.
example_data %>% autoscore::autoscore(acceptable_df = autoscore::acceptable_spellings) %>% as.tibble()
In each of these examples, it is clear that the human and "autoscore" agree the majority of the time. The times that they disagree, it is usually predictably a human error or a subjective judgement that the researcher will have to consider (for example by including alternate spellings of words as we just demonstrated).
Peer-reviewed publications and further tutorials are forthcoming. For more information, contact autoscorehelp@gmail.com.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.