Authors: Qian Wen Chee, Keng Ji Chow, Winston D. Goh, Melvin J. Yap
Categories: Research Article, Social Sciences
Source: PLoS ONE
While a number of tools have been developed for researchers to compute the lexical characteristics of words, extant resources are limited in their useability and functionality. Specifically, some tools require users to have some prior knowledge of some aspects of the applications, and not all tools allow users to specify their own corpora. Additionally, current tools are also limited in terms of the range of metrics that they can compute. To address these methodological gaps, this article introduces LexiCAL, a fast, simple, and intuitive calculator for lexical variables. Specifically, LexiCAL is a standalone executable that provides options for users to calculate a range of theoretically influential surface, orthographic, phonological, and phonographic metrics for any alphabetic language, using any user-specified input, corpus file, and phonetic system. LexiCAL also comes with a set of well-documented Python scripts for each metric, that can be reproduced and/or modified for other research purposes.
In psycholinguistic research, large databases of words, such as the English Lexicon Project (ELP [1]), the CELEX lexical database [2], the Hoosier Mental Lexicon (HML [3]), the MRC psycholinguistic database [4], and the Auditory English Lexicon Project (AELP [5]), have been essential resources for researchers who require data for a multitude of lexical characteristics. These databases typically consist of a comprehensive set of words and their corresponding lexical properties (e.g., word length, number of syllables, pronunciation, etc.), which allow researchers to generate stimuli selection for experimental design (e.g., selecting words that vary on some dimension while matched on others). Such databases of word properties are nicely complemented by megastudies, in which researchers collect behavioral data (e.g., lexical decision performance) for very large sets of stimuli which are defined by the language rather than by a limited set of criteria [6]. For example, the ELP contains lexical decision and speeded pronunciation performance for over 40,000 monosyllabic and multisyllabic words. The powerful combination of normative databases and megastudies allow researchers to evaluate the influence of various lexical properties on task performance.
However, even though many of these databases are comprehensive and have been made widely available, they are associated with methodological limitations. Most critically, researchers are not able to obtain data for stimuli that are not present in these databases, such as less common words, or stimuli that do not form actual words. For example, in lexical decision paradigms, wherein participants are required to distinguish between actual words and made-up words (e.g., ‘murp’), nonwords are necessary as part of the experimental stimuli. Some researchers are also interested in creating specific stimuli, such as pseudohomophones (e.g., ‘brane’) and illegal nonwords (e.g., ‘btese’), to examine the mechanisms underlying word recognition [7]. These types of stimuli are not typically represented in existing databases, and thus, manipulating and controlling for the lexical characteristics of these stimuli can be difficult without some programming expertise.
In order to overcome the above limitation, a number of tools have been developed to help researchers compute various lexical properties for any stimuli. These tools range from online calculators [8–10] to downloadable programs (e.g., N-watch [11]) and R packages (e.g., ‘LexFindR’ [12]; ‘LexOPS’ [13]; ‘vwr’ [14]), and provide options for researchers to compute specific lexical variables for any user-specified input (both actual words and made-up letter strings).
However, despite the growing number of tools that have been created to aid researchers in obtaining lexical properties for any stimuli, existing tools are still largely limited in terms of their user-friendliness and functionality. First, some tools require users to have some prior knowledge of certain aspects of the respective applications. For instance, using R packages requires basic understanding of the syntax used in R, and would be difficult if users are not familiar with any programming language. Second, some of these tools only recognize certain phonetic systems (i.e., symbols to represent speech sounds in pronunciations) for stimuli input. For example, both the phonotactic probability calculators in [8] and [10] only recognize the computer-readable Klattese phonetic symbols [10]. Third, some of these tools only provide for the calculation of a specific lexical variable, so researchers would still have to find some way to obtain data for other variables of interest.
Furthermore, while some of these tools allow users to specify their own corpora that the calculations are based on (e.g., N-watch [11]; ‘vwr’ [14]), most of them still use built-in corpora from existing databases. For example, the online calculator in [10] computes phonotactic probability with respect to the words in the HML [3], while the online calculator in [9] makes computations based on the child corpora of American English [15,16]. While using fixed corpora is convenient, they may not always be representative of the vocabulary of the study population. Specifically, fixed corpora are meant to approximate the lexicon of a typical participant; for instance, the ELP [1] estimates the size of the adult lexicon to be about 40,481 different words, while [15] estimates the size of the lexicon of kindergarten children to be about 3,728 different words. These estimates may not be applicable to other populations (e.g., children of different ages, elderly subjects, subjects with neuropsychological impairments, etc.), which may be of interest to some researchers [17,18]. Computing lexical properties with respect to these corpora may therefore result in inaccurate estimates, but unfortunately, most of these tools do not allow users to specify the size and contents of the corpora.
To improve on current methods, this article introduces LexiCAL (lexical calculator), a standalone executable program that serves as a calculator for a wide range of lexical variables. LexiCAL provides options for users to calculate surface, orthographic, phonological, and phonographic metrics, using inbuilt algorithms. These algorithms were also used to compute the metrics in the AELP [5]. The range of metrics that LexiCAL can compute have been shown to be theoretically important predictors of both visual and spoken word recognition [5,19,20]. Importantly, LexiCAL offers users the flexibility of performing calculations for any user-specified input, with reference to any user-specified corpus and phonetic system. Since the algorithms in LexiCAL are generic, LexiCAL allows users to compute metrics for any letter string (both words and nonwords) in any language with an alphabetic script. With a simple and intuitive user interface, LexiCAL provides maximum functionality and useability. Fig 1 shows the main window of LexiCAL.
Fig 1 Main window of the application.
The program (LexiCAL.exe), along with the Python scripts used to compile the executable, can be downloaded as supplementary material with this article. To ensure continued availability of the program, the latest version of LexiCAL is also available at https://osf.io/ydh6e/ and https://inetapps.nus.edu.sg/aelp/other-resources. Each download will come with the license for LexiCAL, and several open-source components are also listed with the license information.
LexiCAL is available only for the 64-bit version of the Windows operating system, and is not compatible with the 32-bit version of Windows, nor the macOS or Linux operating systems. The program is a standalone executable (LexiCAL.exe) that does not require any installation to launch.
LexiCAL computes and returns the output for each target word in the user-specified input (the input file), based on the user-specified corpus (the corpus file) and phonetic system.
LexiCAL reads target words from a user-specified input file. Users should save target words in a single CSV UTF-8 (comma delimited) file on their machine, without any column headers, and direct LexiCAL to the input file by using the ‘Browse’ option. Orthographic forms (i.e., spelling) should be listed in the first column, and corresponding pronunciations, if applicable, should be listed in the second column. The preview window in LexiCAL will show a preview of the first 20 items in the input file once it has been selected. Fig 2 shows an example of the input file.
Fig 2 Example of a CSV UTF-8 (comma delimited) input file.Orthographic forms are listed in the first column, and corresponding pronunciations (if applicable) are listed in the second column.
As with the input file, LexiCAL reads corpus data from a user-specified corpus file. Users should save corpus data in a single CSV UTF-8 (comma delimited) file on their machine, without any column headers, and direct LexiCAL to the corpus file by using the ‘Browse’ option. Orthographic forms should be listed in the first column, and corresponding pronunciations, if applicable, should be listed in the adjacent column. Word frequencies should be listed in the final column. The preview window in LexiCAL will show a preview of the first 20 items in the corpus file once it has been selected. Fig 3 shows an example of the corpus file.
Fig 3 Example of a CSV UTF-8 (comma delimited) corpus file.Orthographic forms are listed in the first column, and corresponding pronunciations (if applicable) are listed in the adjacent column. Word frequencies are listed in the final column.
LexiCAL recognizes pre-defined phonologies for English (US), English (UK), French, Spanish, Dutch, and German in two phonetic International Phonetic Alphabet (IPA) and Speech Assessment Methods Phonetic Alphabet (SAMPA [21]). Additionally, for English (US), LexiCAL also recognizes Klattese symbols [10]. The IPA consists of a standardized series of phonetic symbols designed to represent all the sounds in the world’s languages, while SAMPA and Klattese are alternative computer-readable phonetic scripts using ASCII characters. Table 1 provides the list of phonetic symbols for each language that LexiCAL recognizes, for each phonetic system.
Other than the built-in phonetic systems, users can also choose to specify their own phonetic system by listing the phonetic symbols in a single CSV UTF-8 (comma delimited) file on their machine, without any column headers, and directing LexiCAL to the file by using the ‘Browse’ option. Phonetic symbols for consonants should be listed in the first column, and vowels in the second column. LexiCAL will recognize diphthongs (combinations of two vowels) as a single vowel if vowel combinations are listed as a single entity. The stress mark for primary stress should be listed in the third column. Fig 4 shows an example of a user-specified phonetic system file.
Fig 4 Example of a CSV UTF-8 (comma delimited) phonetic system file.Consonants are listed in the first column, and vowels are listed in the second column. The application will recognize diphthongs (combinations of two vowels) if they are listed as a single entity. The stress marking for primary stress is listed in the third column.
A list of the surface, orthographic, phonological, and phonographic metrics that are available in LexiCAL, including their descriptions and the algorithms used for calculations, are provided in Table 2.
When computing any metric that involves a target word’s neighbours (i.e., neighbourhood density, neighbourhood frequency, Levenshtein distance-20, spread, uniqueness point, and clustering coefficient) LexiCAL will exclude the target word’s homographs (for orthographic metrics) and/or homophones (for phonological metrics) in the corpus from calculation. By way of illustration, the phonological neighbourhood of the target word “pail” (IPA: /peɪl/) will not include the word “pale” (IPA: /peɪl/), because it is a homophone of the target word. Some databases, such as the ELP [1], provides users the option of including the target word’s homophones in the neighbourhood count, but LexiCAL treats instances of the same spelling and/or pronunciation as a single representation in the lexicon.
By extension, homographs and/or homophones amongst the target word’s neighbours are also counted only once. For example, if the phonological neighbourhood of the word “sail” (IPA: /seɪl/) includes both the words “mail” (IPA: /meɪl/) and “male” (IPA: /meɪl/), LexiCAL will treat “mail/male” as a single phonological neighbour.
For any computation of frequencies involving a target word’s neighbours (i.e., neighbourhood frequency), the word frequencies of homographs (for orthographic metrics) and/or homophones (for phonological metrics) in the target word’s neighbourhood will be summed to give the combined frequency for that particular orthographic and/or phonological neighbour. With reference to earlier examples, the word frequency for “mail/male”, as a phonological neighbour of “sail”, will be the summed frequency of “mail” and “male”.
Once the input file, corpus file, and phonetic system have been selected, users should specify whether their input and corpus files contain only orthographic forms, or both orthographic forms and pronunciations. If the input and corpus files contain only orthographic forms, users will not be allowed to select surface, phonological, and phonographic metrics for calculation. If no stress mark has been specified in the phonetic system, surface metrics would not be returned.
Users can then select variables to be calculated by checking the respective options on the interface. The ‘Default’ button will automatically select number of phonemes, number of syllables, orthographic and phonological neighbourhood densities and frequencies (substitutions, additions, and deletions), and orthographic and phonological Levenshtein distance-20 to be calculated. The ‘Clear’ button will clear all selections.
Users can then press ‘Run’ to begin calculating the metrics, which will prompt users to specify the output file name and location on the directory for the output file to be saved. Progress will be reflected on the progress bar, and the output data will be saved as a single CSV UTF-8 (comma delimited) file. The column headings for each metric in the output file is listed in Table 3. If the application cannot run, an error message will be shown. A list of error messages and their solutions are listed in Table 4. To cancel any calculations already in progress, users can press the ‘Abort’ button.
At any one time, only one instance of the LexiCAL can be open. This prevents users from running multiple instances of the program concurrently.
In addition to the program, each download also comes with a set of 23 Python scripts that were used to compile the executable. The Python scripts are organized by each metric and function, and are well-documented along with comments on code and algorithmic descriptions. Users who are familiar with Python can therefore also choose to calculate the metrics by running individual Python modules on any operating system or platform that supports Python.
Users can also reproduce and/or modify these Python scripts to develop new metrics, or transform existing ones (e.g., adding a logarithmic transformation). The Python scripts can also be combined with other Python libraries to test new theories, or develop new research tools. Any redistribution and/or modification of the Python scripts, however, should be in accordance with the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or any later version (see further licensing information at the end of this article). Anyone reproducing any part of the source code, with or without modification, should also acknowledge and cite this article.
In order to establish the validity of the data calculated by LexiCAL, various lexical properties for the words in the HML (n = 19,321) [3] and the restricted ELP (n = 40,481) [1] were computed and then compared with the data provided by these databases. Table 5 presents the correlation coefficients and between the data computed by LexiCAL, and the data from the HML and the restricted ELP, while the scatterplots are presented in Figs 5 and 6. Despite the slight differences in the algorithms used (with respect to the treatment of homographs and homophones in the corpus), the scatterplots and correlations indicate that LexiCAL’s computations produce values that correspond very closely with these databases.
Fig 5 Scatterplots displaying the relationships between the data computed by LexiCAL, and the data from the Hoosier Mental Lexicon.^a^ substitutions only; ^b^ substitutions, additions, and deletions.
Fig 6 Scatterplots displaying the relationships between the data computed by LexiCAL, and the data from the English Lexicon Project.Scatterplots for word length, orthographic neighbourhood density, and phonographic neighbourhood density are not presented because the data were identical.
LexiCAL is a simple and intuitive application that improves on existing resources by offering researchers the flexibility of computing lexical variables for any stimuli with respect to any corpus of text, using any phonetic system (if applicable). Notably, the lexical variables that LexiCAL can compute include a broad range of inbuilt surface, orthographic, phonological, and phonographic metrics. From a methodological perspective, the program is a useful tool for researchers interested in calculating the lexical properties for any stimuli (both words and nonwords) with reference to any corpus, and should hopefully serve as a resource for experimental design, data analyses, and/or database creation.
LexiCAL is a free software that can be redistributed and/or modified under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or any later version. LexiCAL is distributed in the hope that it will be useful, but without any warranty; without even the implied warranty of merchantability or fitness for a particular purpose. See the GNU General Public License for more details.
LexiCAL makes use of open-source components. The source code for the open-source projects, along with their license information, is provided in the supplementary material.
The program and Python scripts are available with the Supporting Information file. The program and Python scripts can also be downloaded from https://osf.io/ydh6e/.
This research was supported by Singapore Ministry of Education Academic Research Fund Tier-2 Grant MOE2016-T2-2-079 awarded to W.D.G. and M.J.Y. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
The program and Python scripts are available with the Supporting Information file. The program and Python scripts can also be downloaded from https://osf.io/ydh6e/.