3.1.1 Accounts and registration

The majority of language archives we surveyed have materials that are available for download for free and with minimal registration requirements. Many archives appropriately have access restrictions for collections to respect the wishes of language communities and researchers (Nathan 2010). Five archives, including the Repository and Workspace for Austroasiatic Intangible Heritage (RWAAI),¹⁵ required registration to access any materials at all, including the catalog; other archives had a public-facing catalog, even if registration was required for download. Four archives, including ELAR and the DOBES Archive, had multiple tiers of access, where some tiers required registration and/or permission of the depositor for listening or download, while other tiers were unrestricted. One archive, the CLA, has materials that are closed-access, in that they are not available online and must be accessed in person. Two archives, Kaipuleohone¹⁶ and LIA Sápmi (Sami Speech Corpus),¹⁷ restrict all or most of their contents specifically to academic institutions and institution-affiliated researchers, a limitation that may exclude members of language communities. Others restrict only parts of their materials to those affiliated with academic institutions. The CHILDES Data Repository¹⁸ includes password-protected collections available only to faculty members, and the CLARIN Slovenian Repository¹⁹ requires that users access certain materials restricted for “academic use” through their institutional emails. While there are good reasons why collections may be not freely available, some of the convoluted, unclear, or heavily outdated procedures for requesting permission could be fixed. Archives that do not streamline the permission forms or provide unclear contact information could be updated. For example, account registrations requiring manual approvals, or emailing specific individuals, should be automated.

While the majority of archives are entirely free for use, and we did not encounter any archives²⁰ requiring payment of fees to access collections during our audit, we acknowledge that for some researchers, particularly those who do not have institutional membership with archives or sufficient funding, the cost of accessing an archive may be prohibitive. The Linguistic Data Consortium (LDC)²¹ and the European Language Resources Association (ELRA)²² are two examples of archives requiring payment in return for access to materials; these fees can range into tens of thousands of dollars (Vann 2006). Endangered language archives have tended towards a model where the archive is supported through institutional or grant funds, with costs supplemented by fees from depositors (similar to “gold” open access models for academic publication). Some archives have recently requested that depositors include archiving charges in research grant applications. Clearly the funding model for ongoing support for endangered language archives needs to be investigated in more detail.

3.1.2 Display language

The language(s) used to display metadata and to navigate the site may also limit accessibility of the materials. Bias towards English-language users and lack of built-in site translations disadvantages researchers whose primary language is not English and may prevent community members from accessing documentation of their own languages or other languages which they regularly use.²³ While the arrival of digital media devices and technologies can facilitate the creation of a “social network of digital exchange” of cultural heritage for Indigenous communities (Mansfield 2014:66), unavailability of these resources in endangered languages further entrenches generational and educational divides in language communities where acquisition of literacy, particularly in English, is not widespread. A number of linguists and Indigenous community members have expressed concern that “the majority of digital resources available to Indigenous users are in English, even though English is not a first language for many” (Carew et al. 2015:310). Only 14 of the archives we examined provide more than one language interface, and not all of these had fully functional language options. We point to PARADISEC²⁴ (see Figure 1) as an example of an archive providing information through 7 regional languages (though unfortunately not on the mobile site).

Archives that focused on languages of a particular region often provided interfaces relevant to their users. For example, AILLA has interface options in Spanish, and PanGloss is fully implemented in both English and French.²⁵ ELAR and CLA have interfaces only in English, though at the collection level, ELAR allows filenames and metadata to be in other languages and scripts, which helps users if they know of the collection.

We applied Google Translate to the exclusively English archives (testing languages such as Korean, Uzbek, Kyrgyz, and French). Translations were inconsistent, incomplete, and sometimes misleading. Some localizations translated only parts of the site text, leaving others, such as an embedded map and filenames, in English (see Figures 2, 3, and 4). Therefore using Google Translate as a workaround for untranslated sites is not a straightforward alternative.

It is also worth noting that when site translations are available, options are predominantly languages of European origin. This is especially striking given the scope of archived languages, most of which are indigenous to Africa, Asia, and the Americas. Lack of translations into major regional languages limits the abilities of scholars to use these archives, creating a bias in the demographics of researchers and restricting potential scholarly innovation. For non-Indo-European languages whose structures differ greatly from that of languages like English, French, and Spanish, automatic translation programs like Google Translate and Yandex are especially prone to offering confusing and poor-quality translations. We recognize that this is a much bigger problem than what individual archives can solve. For example, search engines filter out search queries in languages other than the query language (which made it almost impossible for us to search for archives outside the anglosphere internet).²⁶ However, at the collection level, depositors should be encouraged to provide materials in languages that will be most usable for community members, and the substantial additional time costs for doing so should be recognized explicitly.

3.1.3 Disability accommodations

We acknowledge that disability accommodation remains a critical, and often-overlooked element of archive accessibility, and indeed, the accessibility of any digital material. In regards to the structure of websites and storage of data such as text files, it is essential that Web content – including archives – is presented in a way that is accessible for visually-impaired researchers.²⁷

It is generally agreed as a principle of accessible Web design not to make different elements of a site distinguishable only by their color (Campbell 2018). In order to assess color blindness accessibility, we put each archive through a filter²⁸ mimicking how each site would look to users with 3 of the most common kinds of color blindness. Sites were subjectively reviewed by a member of the team who is colorblind. The archives we surveyed, largely, performed well in this regard. The main issue raised by our survey was the low contrast between font and background colors, which may compromise readability for users with certain kinds of color blindness and other visual impairments; it may also inconvenience users with certain color and brightness settings on their computers and browsers. The websites for SIL’s International Language & Culture Archives and the Rosetta Project revealed such problems.

3.1.4 Recommendations

The restrictions archives place on access to language data are there for a reason; however, it is important that these restrictions do not place too much of a burden on researchers and language communities looking to access their contents. Therefore, we suggest archives streamline the process of requesting access permission. More specifically, we recommend that request forms be built into the site itself, with additional capacities for automated password retrieval. This is especially critical for those archives (such as ELAR) whose code is not built for long-term accessibility, as passwords cannot be reset and the application permission form is built on Google Surveys. For archives not already implemented in multiple languages, we strongly suggest expanding the scope of display languages offered, especially those languages which may be relevant to language communities and local researchers. Furthermore, we recommend against applications that must be physically posted to the archive, given their inefficiency and potential to disadvantage researchers in areas underserved by the postal system.²⁹ Following the principles of accessible Web design will make great strides in overcoming barriers for researchers who require assistive technology. Even in the absence of laws like the American Disabilities Act (1990) (or varying legal requirements across different countries), it is important and not too difficult to improve what is already there.

3.2.1 Search functions and mislabeling

Search functions are vital for navigating large collections, but they can be made frustratingly slow and even useless depending on their available options. Six archives offer a map search function, which allows users to browse collections by location. This function is especially useful for more casual users or for those who are not searching for specific data, but it presents its own challenges. Archives like ELAR and the California Language Archive use Google Earth, and many others use similar platforms. The California Language Archive does not outwardly indicate whether each collection on the map is available online. This makes it initially seem as if there are more resources readily available to users than there actually are. While these issues are a cause for frustration, they are not necessarily debilitating, and the map function tends to be a useful visual aid for users. We point to PanGloss as an example of the map function at its most useful; its map function is easy to navigate, contains information on the title, researchers, and types of resources available for each collection (as well as a link to each collection), and can be filtered by the criteria “with annotation” and “with video”. In contrast, AILLA’s map function is non-interactive. That said, some location information within PanGloss was inaccurate.

Lack of transparency about the contents of collections was observed in numerous archives. Users may have little information about what a deposit contains before accessing its contents. Researchers often have specific criteria in their search for language materials—for example, linguists looking to perform certain kinds of phonological analysis may have a preference for collections whose video and audio recordings are all fully transcribed, and rule out collections with too few hours of recorded material or those that consist only of written materials. Other criteria may include the specific dialect(s) documented, date of creation, file type, number, age and location of speakers, or specific individuals. Some of these categories can be aggregated automatically, while others require manual labeling. While the lack of some of this information is due to incomplete metadata provided by depositors, we encourage archives to make such information easy to find. ELAR, for example, includes a collection landing page consisting of sections for “summary of the deposit”, “groups represented”, “language information”, “special characteristics”, and “deposit content”, though the quality and specificity of the information in these descriptions varied greatly between collections. This could be a point of evaluation for individual collections. In most collections, the metadata about the holdings is a file within the general collection. It is not consistently named and where collections have many files it is difficult to find. Archives could assist the retrieval of such information by flagging such metadata files directly or including an explicit link to the metadata file(s) within the collection overview.

Most archive portals include search bars, but these have varying degrees of usability. One important feature is a filter function, especially for larger archives. All but seven of the archives we investigated have some kind of search filter function. Some filtering options include language, speaker, depositor, file type, topic, and country, among others. However, the availability and usability of the filter function was inconsistent. ELAR’s search filter options vary by collection, while The African Language Materials Archive,³⁰ Digital Himalaya,³¹ and AILLA all lack a search filter function entirely, making large collections more difficult to search. PARADISEC had a flexible search and filtering interface, at the item or collection level.

A useful search feature available in some archives is the ability to search within collections. This feature is especially useful, almost necessary for archives that contain large collections. However, despite its importance, we only found the feature in four of the archives we examined. Such free-text search increases finding options for collections extensively, allowing more refined searches than filters alone. For example, a filter may exist to restrict files to .xml, but a test search makes it easier to distinguish between Flextext transcripts, ELAN transcripts, and .xml-format metadata. These are all .xml format files but have very different functions. At the collection level, searches were hampered by missing metadata, incorrect tags, case sensitive searches and inconsistent metadata (e.g., searches returning either audio or Audio as filetype), empty folders, and broken URLs within collections. Correcting these small but time-consuming errors would improve intra-archive searches.

Two of the most useful search filter categories are media type and file type (see Figures 5 and 6). Many researchers using these digital archives can only use files of a specific media type (e.g., videos or sound recordings), or, in cases where they plan to use certain software in their research, certain file types (e.g., .pdf or .wav files). File type and media type filters greatly reduce the time a researcher must spend browsing files to find what they need. Despite this importance, only five of the archives we looked at offer the option to filter by file type, and one of these archives, The Language Commons,³² returns files that aren’t bundled (with related files of different file types) when the file type filter is employed, causing users to miss potentially useful materials. Similarly, only four of the archives we looked at offer the option to filter by media type. Even fewer allowed users to filter by specific file extensions (such as .mp3 or .wav), and, when offered, the archive often displayed results with mislabeled extensions (.xml for .eaf, for example).

Mislabeled file types are another issue we encountered. ELAR and AILLA, for example, rename ELAN³³ .eaf files and FLEx³⁴ .flextext files as .xml (see Figure 7). While these are underlyingly XML files and alternate extensions are visible upon downloading the files, one needs to know how to change the file extensions in order to open the files with the appropriate applications. It is also difficult to differentiate ELAN audio transcripts from FLEx dictionary or interlinearized texts, which are both listed as .xml files but have different underlying data structures.

3.2.2 Metadata

We noted considerable inconsistency in what type of metadata was available, across archives and collections. It is easy for relevant files to be lost in a search because they do not have the type of metadata used in the search.

Another issue we discovered was the use of different layers of metadata. In many cases, important metadata was hidden inside the folders of a collection, making it difficult for a user to find the specific information they need. AILLA, for example, has three layers of metadata: one layer is for the whole collection, another layer is found within each individual folder within each collection, and the final layer is attached to the individual files themselves. Such layering, combined with the frequent gaps in available metadata, makes it extremely difficult to find desired information and reduces the accuracy of the search function.

Sullivant (2020) provides detailed recommendations for collection metadata, breaking down these recommendations into categories based on importance. We point to The California Language Archive and PARADISEC as two archives that do a good job of including “first priority collection metadata”. Finally, it is important to note that, while many of the archives we examined do include the most important information in their collection metadata, almost none include the information in Sullivant’s next two tiers of recommended metadata. While archives are reliant on the metadata provided by depositors, this only reinforces the points made by Sullivant (2020) and Burke and Zavalina (2020) that metadata is crucial to the usability of a collection. The DACS³⁵ standards may also be useful for both depositors and archives to introduce and maintain consistency.

3.2.3 Site maintenance

Other issues impeded discoverability, with archives being incompatible with specific browsers, requiring defunct software, or failing to load entirely. This occasionally varied depending on the individual user in ways we were unable to solve. For example, three of the team members found that the APS Digital Library³⁶ would not open for them unless they accessed it via Yale University’s VPN, while the remaining team members could access the site with no difficulties from off campus, all using recent versions of Chrome on MacOS 11.6 or Windows 10.

Six of the 41 archives gave web access errors or were unreachable.³⁷ Some, such as ALORA,³⁸ could only be accessed with the Wayback Machine.³⁹ While these workarounds do allow users to access materials, users who are unfamiliar with the Wayback Machine would be deterred from retrieving relevant information. Moreover, the Wayback machine may provide access to the catalog, but not the files in the collection itself. Links provided within archives often faced the same issues, defeating the purpose of being an archive that safeguards data.⁴⁰ Furthermore, at least two archives⁴¹ still required some use of Adobe Flash Player (see Figure 8), which was phased out by December 2020.

Many archives contained broken links, though this differed in extent and severity. The problems related to both site-internal and -external links, and problems arose due to internal site reconfigurations (such as those of the British Library’s Endangered Archives Collections)⁴² as well as the removal of individual pages. It would be ideal for archives to not rely on external links, but when necessary, regularly checking for outdated links is crucial. Broken links not only hinder the usability of archival materials from an end-user perspective, they also hinder the discoverability of such webpages. Search engines penalize broken links⁴³ in search results, thus making archive sites with such links less findable. As language resources are entrusted in archive sites’ stewardship, it is important that they remain discoverable by those who wish to access these language materials.

Corrado and Sandy (2017) draw attention to the lifecycle of a project, as defined by the Life Cycle Information for E-Literature.⁴⁴ They argue that “institutional commitment…ensuring that enough financial resources are available to sustain the initiative” is necessary for digital preservation to be successful (Corrado & Sandy 2017:11). In order for stewardship organizations to faithfully fulfill their fiduciary duties as language resource stewards, website maintenance must receive ongoing support to keep up with rapidly-changing software and security compatibility requirements.

3.2.4 Recommendations

Offering more detailed descriptions of a collection’s contents, specifically, media types (video, audio, text, etc.), completion state of any transcriptions or translations, and number of hours of recorded material, would help researchers evaluate the utility of a collection for a particular purpose, and give community members a sense of what is in the collection. Allowing searches by file type would allow researchers to further refine their queries and determine the usability of a given collection for their research purposes; we also recommend that archives correctly label file types and remove filetype capitalization dependencies on searching.⁴⁵ We also suggest that archives make it clear to depositors what types of information are indexed for searching, and how researchers can structure their collections to make them usable. To make archives more easily discoverable, we recommend archive managers use the Sitemaps⁴⁶ protocol set to provide site-internal content information to search engines. Finally, we suggest that depositors consider how they use links to external sites in their deposits, archiving copies where appropriate (or pointing links to the Internet Archive). We suggest that archives regularly check for link breaks (e.g., by using automated checking tools that generate reports, such as the Broken Link Checker plugin),⁴⁷ particularly to archive-internal pages.

3.3.1 Site content, structure, and downloads

The available content and structures of archive sites posed the first issue with functionality. We note that some of the following concerns are affected by the choice of CMS of individual archives. We attempted to track CMS use across archives, such as whether the archive used a common CMS such as Mukurtu,⁴⁸ DSpace,⁴⁹ or a bespoke platform. However, information about the CMSs underlying the archives in our audit was not easily accessible; an overwhelming majority of archives had no publicly available information at all about their CMS. Half of the archives mentioned the institutions that supported the development of the archive, or external servers where related language corpora were hosted, but the information about infrastructure was not available for enough archives for us to track it. We acknowledge, however, that site structure and content capabilities are closely linked to choice of CMS.

The sites examined here vary extensively in their holdings and scope. Some sites labeled as archives only hosted one or two resources (Magoria Books Carib & Romani Archive),⁵⁰ which sometimes required purchase, while others hosted none at all (Multimodal Learning Corpus Exchange).⁵¹ Others, such as the SIL International Language & Culture Archives,⁵² appeared to function more as directories with both links to external resources and hosted materials. They were not “archives” in the sense of storing and safeguarding materials. This is in contrast to archives such as the ELAR archive, which has full hosting and offers (per R3data.org) more than 462,048 results.⁵³ The most prevalent issue impeding archives’ functionality was the lack of a bulk download option. The vast majority (34/41) had no bulk download option for either text or audio/video. Two⁵⁴ had bulk download options for text files only, and five⁵⁵ provided download links for zip files containing all or a selection of the files in the corpus. Requiring users to download files individually not only results in loss of time, but also renders some collections (e.g., those with 15,000 audio files) virtually inaccessible because of the sheer number of clicks, ranging from 1 to 7 per file, required to download their contents. Further, when individual downloads are the only option, users would benefit from knowing exactly how many files are in each collection, allowing them to assess their own storage capacity before attempting to download a corpus.

Another concern that results from downloading files individually is the loss of arrangement of items within a collection. For example, nested files lose their relationships to each other and must be manually re-sorted when downloaded onto a drive. This is assuming that the archive site has not already collapsed structures that existed when researchers originally deposited their files. When this happens, crucial information can be lost for collections that depend on file structure to match transcripts and metadata files to audio and video files (for further discussion of arrangement, see Patterson 2021: §6.3.1).

We do recognize that there are non-trivial issues concerning bandwidth, web server traffic, and validation of large files that limit download capabilities and may require additional funding to resolve. Still, since these issues directly affect archives’ functionality, they should be addressed sooner rather than later. Even if downloads must be done individually, solutions such as putting all of a collection’s download links on a single page (as opposed to requiring users to enter into individual folders to download) exist. We draw attention to the DOBES Archive for providing an effortless method of downloading files in bulk. Their “basket” system allows users to select and bundle individual files or entire collections, then after an amount of time proportional to the number of files they have requested, a link to a zip file is emailed directly to them.

Other issues surrounding downloads included non-functioning download buttons or downloads that resulted in unreadable data. AILLA’s download links are blocked by Chrome and Firefox browsers due to security settings, and could only be accessed by changing web browsers. The Hindu-Kush Areal Typology,⁵⁶ while not strictly an archive, had a bulk download option for wordlists. However, users had to ensure that they were properly opening the UTF-8 encoded CSV file in order to read the data without broken text. While workarounds like these exist, they may deter users with less familiarity with technology from using such archives effectively.

3.3.2 Recommendations

Firstly, and perhaps most critically, we suggest adding the option to download files in bulk, including an option for the entire corpus and for each folder in it, while preserving the original arrangement configuration. We recognize that this may be a complex request, given how file storage may work for the archive, but it is a necessary part of making files accessible. A 15,000-item collection with no bulk download option is neither accessible nor realistically usable. Furthermore, we suggest archives either allow depositors to preserve the original file structure of their collections upon deposit, or develop tools to help them better structure collections once archived in-site, for example through tags. It is vital that archives provide layout guides and naming conventions for depositors, so that users may quickly locate corresponding files and recreate file structures in the event that they are lost, and care should be taken when depositing collections to make sure that vital information about metadata and collection structure is not lost.