vol. 25 no. 4, December, 2020

Proceedings of ISIC: the information behaviour conference, Pretoria, South Africa, 28th September to 1st October, 2020.

Poster session

The capability of search tools to retrieve words with specific properties from large text collections

Liezl Ball, and Theo Bothma.

Introduction. With the increase in the availability of digital text collections for humanities researchers, tools to enable enhanced retrieval are required. If words with very specific properties could be retrieved from a text collection more accurate linguistic and other analyses can be made. There are a range of properties and metadata that could be specified for retrieval, from morphological data up to bibliographic data. Furthermore, the bibliographic data should not only be on item level but extended to the text-level. For example, in an anthology each section could be encoded with the author of that section. Such extended metadata will enable fine-grained retrieval.
Method. In this study, current tools were evaluated to determine to what extent they allow users to retrieve words with specific properties from a text collection.
Analysis. The analysis is limited to the following criteria: interface design, metadata, search options, filtering and search results.
Results. Currently, it is not possible for a user to retrieve words with specific properties from a text collection.
Conclusion. An extended set of metadata should be used to encode text to enable retrieval of words on a fine-grained level.

DOI: https://doi.org/10.47989/irisic2030


The amount of digital text data available is increasing rapidly. Researchers from different disciplines are interested in using these data in their research. Though one might associate the use of large amounts of data with the natural sciences, researchers in the humanities are increasingly interested in the use of large digital collections for research. Some of the well-known large digital text collections include Google Books, Internet Archive and HathiTrust.

The availability of such data, and technology to process the data, are opening up new possibilities in the field of humanities (e.g., Howard, 2017; Nicholson, 2013; Rydberg-Cox et al., 2000). Some tools have been developed to help users to search for information in large digital text collections, for example the Google Books Ngram Viewer (https://books.google.com/ngrams). Using this tool, a user can see the usage frequency of a term over a period of time. Some interesting research using this tool has been done (e.g., Acerbi et al., 2013; Michel et al., 2011; Ophir, 2016).

Problem statement

Though there are some exciting developments, there are some criticism against current tools and methods. For example, Google Books Ngram Viewer has been severely criticised for the lack of metadata (e.g., Koplenig, 2017). Furthermore, there are some additional features that could enhance the tools, specifically in terms of metadata. If texts are encoded with detailed metadata more powerful searches will be possible. Two examples will suffice. Morphological metadata could be added to enable a user to search for specific inflected forms of irregular verbs; or metadata could be used to indicate structures in a text. For example, if a document contains multiple languages, each section could be encoded with tags to specify the language of that section; this will enable a user to search for words in a specific language on a very detailed level. This means that texts need to be encoded on different levels, from metadata to indicate structure in texts to detailed morphological data.

This study forms part of a larger research project to determine how texts could be encoded on a detailed level to improve retrieval of words or phrases with specific properties from large digital text collections.

The first step in answering the broader question will be to examine the most popular tools used currently for the retrieval of words from a large text collection and answer the following question:

To what extent do current tools allow a user to retrieve words or phrases with specific properties?

Significance of the topic

This research makes an important contribution in the field of information seeking and use. The overwhelming amount of text data available will only be effectively utilised if useful tools are developed that enable users to retrieve the relevant information in terms of their information needs. This study focuses on the ability of users to retrieve words with specific properties from a text collection. By evaluating current tools, recommendations can be made for further development of advanced search tools used on large digital text collections. Improved tools can help researchers, authors, and other users to observe trends and analyse the results.

Examination of tools

Six tools that can be used to search for words (or phrases) in a digital text collection (corpus) are examined to determine to what extent a user can search for words or sections of texts with specific properties. The six tools that have been identified are the Google Books Ngram Viewer, HathiTrust+Bookworm, Perseus Project, Voyant Tools, TXM and BNCweb. These tools are used to search in large digital text collections. These tools are examined according to the following criteria: interface design, metadata, search options, filtering and search results. More criteria, such as complexity of use, help files and corpus design, could be considered in future studies. Key features in these categories are highlighted. It is beyond the scope of this paper to offer extensive detail of each tool.

Google Books Ngram Viewer

The Google Books Ngram Viewer shows the relative frequency of words (or phrases) used over a specified period of time. The data used by this tool are from a selection of books from Google Books. An example of a search is shown in Figure 1. In this example the tool is showing all instances of well where it is a noun, of well where it is an adjective and where well modifies done.

Figure 1: Google Books Ngram Viewer

Figure 1: Google Books ngram viewer

Search results: The results are displayed in a graph. The results do not link to the underlying data directly and it is not possible to see examples in context. Below the graph are links to predetermined searches in Google Books for the search terms used in the specific search.


The HathiTrust+Bookworm) also visualises the frequency of words over a period of time, however a user can filter the results using bibliographic metadata. The data used by this tool come from the HathiTrust Digital Library. In Figure 2 a search for the terms carriage and chaise is shown. Filters, such as publication country, class and resources type, have been applied.

Figure 2: HathiTrust+Bookworm

Figure 2: HathiTrust+Bookworm

Search results: The frequency counts are shown on a graph. It is possible to click on a point on the graph and see a list of results for that point. The items in the list of results link to the texts in the digital library. However, it only links to the volume, not the term(s) in context.

Perseus Digital Library

The Perseus Digital Library) was specifically developed to explore the possibilities of digital collections. The focus was originally on Greek and Latin material, but the collection has expanded. Figure 3 shows the results of a search for all forms of the Latin word suis, limited to the Greek and Latin collection.

Figure 3: Perseus Project

Figure 3: Perseus Project

Voyant Tools

Voyant Tools) is a free, online tool for text analysis. Texts to be analysed are added by the user. The main page contains several panels, each with a different tool useful for text analysis (for example to view trends). Figure 4 shows a section of the tool.

Figure 4: Voyant Tools

Figure 4: Voyant Tools


TXM is a free text analysis environment. One of the main aims was to be compatible with encoded texts (for example, texts encoded with Text Encoding Initiative). It is a powerful tool and it is beyond the scope of this study to offer an in-depth description of all the features of this tool. Specific features relevant to this study will be highlighted. Different corpora can be imported and studied. A query in the concordance is shown in figure 5. This example query searches for a sequence of four words, where the first word must be a verb, the second word must be le, the third word must be a noun and the last word in the sequence must end in an -e.

Figure 5: TXM

Figure 5: TXM

Search results: Various types of results can be retrieved. The concordance lists the retrieved instances in context. It is possible to link to more context.

BNCweb (CQP-Edition)

The BNCweb (CQP-Edition)) is one of the tools that are used to explore the British National Corpus (BNC). The interface is simple, while allowing users to enter complex queries based on a corpus query language. The results of a search are shown in Figure 6.

Figure 6: BNCweb (CQP-edition)

Figure 6: BNCweb (CQP-edition)

Search results: The default option is to display retrieved instances in context. It is possible to link to more context and see information about the text the instance is from. Other display options are available.


There is increasing interest in using large digital text collections for research purposes. Several tools have been developed to allow a user to explore such data. These tools allow researchers to do more than what is possible using a general search engine. There are options to visualise trends, make selections based on morphological or syntactic properties, retrieve sections of encoded data or filter according to bibliographic data. However, there are several limitations in the tools currently available. Tools with simpler interfaces do not allow users to filter on a detailed level. Some tools allow very specific filtering but are complex. None of the tools in this study allows a researcher with little prior training or knowledge of encoding (such as XML), to search for words (or phrases) with specific properties in a large collection of texts and ideally view the usage frequency of the word(s) over time. More work should be done to determine what metadata can be used to encode texts to enable meaningful retrieval, and how a tool can use such metadata to support retrieval in an easy and intuitive manner.

About the author

Liezl Ball is a lecturer in the Department of Information Science at the University of Pretoria, South Africa. She completed her master’s degree in 2016 and is currently working on her PhD. Her research is in the field of digital humanities and investigates how large text collections may be enhanced with metadata to improve retrieval. She can be contacted at liezl.ball@up.ac.za
Theo Bothma is Professor Emeritus / contract professor in the Department of Information Science at the University of Pretoria, South Africa. He is the former Head of Department and Chairperson of the School of Information Technology (until his retirement at the end of June 2016). His research focuses primarily on information organisation and retrieval, information literacy and e-lexicography. He can be contacted at theo.bothma@up.ac.za


How to cite this paper

Ball, L., & Bothma, T. (2020). The capability of search tools to retrieve words with specific properties from large text collections. In Proceedings of ISIC, the Information Behaviour Conference, Pretoria, South Africa, 28 September - 1 October, 2020. Information Research, 25(4), paper isic2030. Retrieved from http://InformationR.net/ir/25-4/isic2020/isic2030.html (Archived by the Internet Archive at https://bit.ly/3meU2cA) https://doi.org/10.47989/irisic2030

Check for citations, using Google Scholar