One of the potential advantages of using multimedia applications in education is the possibility of reducing the occurrence of gender bias in instructional material. As computers and computer applications become more prevalent in education, gender bias has been carried over from written text into computer based text. The gender bias detection and correction techniques discussed in this paper have been incorporated into a software tool that can pinpoint instances of gender bias in electronic forms of text and allow users to modify text where appropriate. This paper discusses the development of this tool and offers recommendations for using the tool in multimedia application design.
Multimedia, on the other hand, can escape from the confines of linear transmission. Although many definitions are available for multimedia it has two essential elements in this discussion. First, multimedia is simply the integration of text, graphics, audio, images, animation, and full motion video into a single presentation unit. A multimedia presentation, under this definition alone, could be linear. Second, non-linear navigation through the multimedia material must be possible. This definition of multimedia is often called hypermedia. This term represents the ability to move through information non-sequentially. A hypertext system includes only textual material.
Hypermedia and hypertext are both forms of multimedia and very different from written text because non-linear presentation of information is possible. Multimedia is often said to mirror the way in which people think, learn, and remember by providing the capability to move from words to images to sounds. Analysis, interpretation and in depth exploration of the material is possible.
Interactive multimedia has the potential, therefore, to overcome many of the problems associated with written texts. Written texts that are static, outdated, one dimensional, and biased towards one viewpoint make little contribution to a learner centred environment where knowledge is constructed instead of simply repeated. Traditional shortcomings, of written texts such as gender bias can be alleviated by presenting multiple representations of a subject for a learner to manipulate and integrate.
Numerous studies have shown that gender bias, in a variety of forms, appears in educational materials in just about every subject area (APA Task Force, 1975; Britton, 1974; Gunderson, 1975; Kepner & Koehn, 1977; Kernberger, 1990; Pyle, 1976; Sadker & Sadker, 1979; Scott & Schau, 1985; Warren, 1988). Some gender bias is topical. For example, history text books are often referred to as being written about "dead white males" (Hughes, 1992), since the texts rarely cover the contributions of females or minorities. Other gender bias is role related. For example, casting the role of a doctor as a male and a nurse as a female reinforces stereotypical prejudices.
Studies (Schau & Scott, 1984; Sadker, Sadker & Klein, 1991) have shown that gender biased materials can have adverse affects on the educational process, can perpetrate gender related stereotypes, and can effect the emotional well being of students. Them effects can be greatly reduced and even eliminated through the use of non-biased materials.
Invisibility occurs when a gender, male or female, is not included in the work. Invisibility can only be detected after an entire work has been reviewed. Random sampling of the text is not sufficient to show that a certain gender is not included in the entire work. A common occurrence of invisibility is in history books where authors are accused of not including the roles of women and minorities.
Stereotyping occurs when genders are assigned to their "traditional' roles in the text. For example, men are doctors, women are nurses. Stereotyping detection requires readers to tabulate all the characters and their roles in a body of text. Elementary textbooks are often singled out as examples of stereotyping the roles of males and females.
Imbalance/Selectivity is an imbalance in presentation of the material, by selective interpretation of events being reported. This can lead to a distorted reality. To detect this form of bias, the reader needs to have a broad knowledge of the subject area being analysed. For example, the statement, "Women were given the vote after World War II", is imbalanced because it ignores the work of the suffragettes who fought to earn women the right to vote.
Unreality is similar in nature to imbalance, this is when the author chooses to avoid controversial issues, or reduce large complex issues to a simple, if inaccurate, explanation. Again, this requires knowledge of the subject area. For example, the portrayal of women solely in the roles of housewives., despite that fact that a majority of adult women work outside the home.
Fragmentation/Isolation can occur in an attempt to include women or minorities in their works, authors, editors, and/or publishers, will add additional chapters. "Famous Women Inventors," would be an example of a chapter in a history book. This obvious form (and some not so obvious forms of separate sections and alternating chapters), can be detected when viewing the material as a whole. Again, random sampling can not be used to detect fragmentation/isolation.
Linguistic Bias is the general use of masculine terms and pronouns to reflect a composite audience. This form of bias is sometimes very subtle, since it is a widely accepted writing style. Detecting linguistic bias relatively easy, but it can be overlooked due to cultural background and upbringing. Some obvious forms of this bias is the use of "he" to refer to the entire population, or the use of the suffix "man" in occupations, such as "policeman".
Visual Bias refers to the pictures and illustrations in a work, and the ratio of male to female representation. Aside from inequality, visual images can also contain stereotyping. Due to the difficulty in the automation of image recognition and visual processing, it is almost impossible to automate the detection of this form of bias.
Gonzalez-Suarez and Ekstrom (1989) analysed seven elementary school textbooks as part of an international study of sex stereotyping. Males were present 64% of the time in the text, and 61% of the illustrations. They also determined that males were more likely to be presented in occupational or historical roles, while women were depicted in ways that emphasised their personalities.
Kernberger (1990) discusses the teaching of non-sexist language to learners of English as a Second Language. Arguments both for and against the use of masculine words are presented, but the author concludes that use of "he" and "man" and their derivatives to refer to people of both sexes should be considered inaccurate.
The current manual process for detecting gender bias is slow and difficult, sometimes requiring months of effort by human readers. The effort and time to train skilled readers is also quite extensive. Readers must develop the ability to simultaneously analyse content while evaluating language constructs. Sonic forms of bias can be quite subtle and may elude detection by even the most highly trained and diligent reader. The personal background of the reader may also have an influence on the detection of bias. People from different cultures, even different geographical regions of a culture, may evaluate language in different ways.
As in most tedious tasks, the factor of boredom also comes into play. The task is slow and arduous, requiring extreme concentration on the reader's part. It is not enough to simply read the document. The reader must also look for words, phrases, role models, balance, and meanings, that conceal gender bias. Delicately hidden gender bias can easily be overlooked when the reader tires or becomes bored. A reader can develop a false sense of confidence when a text is primarily non-biased - readers may accelerate the analysis, missing biased references.
Typically, school districts, publishers, authors, and educators, employ experts to read and analyse books for gender bias. Currently, this process can take several months, and the results may not always be reliable. The process of analysing instructional materials for gender bias using human readers is very time and resource consuming, and is not guaranteed to be 100 percent accurate. Readers who detect gender bias have to be trained and typically have an inter-reader reliability rating of 85 percent or better.
Many publishers, such as Scott Foresman, Harper and Row, Ginn, Macmillan, Houghton Mifflin, and J. P. Lippincott (now out of business), have published guidelines for authors to identify and eliminate the sexist portrayals of females in their materials (Britton & Lumpkin, 1976; Weston & Stein, 1978). However, not all research shows that these guidelines are having much impact. Bertilson (1982) indicated that newer textbooks lack any significant improvement. However, various content studies between 1980 and 1990 have had conflicting results (Sadker, Sadker & Klein, 1991).
While this process is being performed in the paper publishing world, there are few similar control mechanisms in the hyper/multimedia publishing arena. Many of the educational software packages are produced in house, or by computer firms, with little or no knowledge of gender bias.
The program, written for the PC family will scan the text looking for gender words from a set of dictionaries. It keep count of the words, and presents a ratio of male to female words. It can also pause at each occurrence, much like a spell checker, and allow the user to examine the use of the gender term in context. The program will also keep track and assign genders to last names, when a first name or a title is used.
During the design of the program, a number of natural language parsing techniques were tried and abandoned. The most obvious problem encountered was the sheer bulk of the material. The program is designed to word with the entire book, not just a sentence, paragraph or a page of text. Most natural language parsers only work on a limited amount of text, and a small vocabulary and sentence style. No program exists that can interpret a robust grammar. In fact, humans sometimes have problems interpreting the meaning of some complex sentences. The major problems facing researchers in Natural Language Processing are summarised by Schutzer (1987):
Bundy (1980) states that, "Even determining which person a pronoun like 'he' refers to is a hard task. The need to consider all possibilities, and to follow trails that may later turn out false, complicates the programming." This can be complicated even further if the pronoun "he" is not referring to an individual, but in fact being used as a generic reference.
The 50,000 word document was processed in batch mode on a 33 MHz 386 machine in about two minutes. Two passes over the text was performed to ensure that all gender references are found. The program revealed that there were 387 male words and 44 male names, while there were only 76 female words and 11 female names. There were the same number of occurrences of proper names, ten each. This represents a ratio of approximately five male references to each female reference.
When the program was run in interactive mode, the reader could analyse the gender occurrences and choose to ignore some if the computer has misinterpreted the text. For example, the word "Stanford" was ignored because in this context it refers to the school, rather than a man's first name. In this mode, the program took 11 minutes to run. Because some words were ignored, a third pass of the text was performed to get a proper tabulation. In this mode the program showed that there were 387 male words and 24 male names, but still only 76 female words and 11 female names. This was still approximately a five to one ratio.
The program also produces a table that can be imported into a spreadsheet and plotted. This can be used to find isolation and fragmentation.
The nature of hypermedia development also enables the course developer to emphasise major topics by guiding the user to central themes through multiple representations of the material, provision of both corroborative and conflicting material and cognitively astute construction of the linkages between the content nodes. In reality, many course developers do this poorly and hypermedia course materials are still often characterised by a lack of planning, organisation and goal identification.
Although enhanced learning has been anticipated as a result of using hypertext and hypermedia systems for instruction, many of these systems have not fully lived up to this expectation. One of the reasons for this shortcoming is that most of these systems have been designed only for knowledge presentation where users simply follow links established by authors or for knowledge representation where efforts are made to provide direct and meaningful indications of the relationships between the information in the nodes and links through such devices as maps or graphical browsers. While the latter is clearly superior for instruction, it still falls short in providing the user the ability to construct new and personalised knowledge in a context rich learning environment such as would be found in a system designed for knowledge construction.
Hypermedia systems designed for knowledge presentation or knowledge representation fall within the realm of an objectivist theory of learning wherein the reality of the author is transmitted to the user in a communication based model. Hypermedia systems designed for knowledge construction view knowledge acquisition as an active process of construction whereby learners come to understand the world in which they live by an ongoing process of making sense out of new information - by creating their own version of reality instead of simply receiving the author's view.
Some of the goals of constructivist learning are to aid the user in learning how to ask relevant questions, how to test their views against alternate views, and to become aware of the knowledge construction process (Cunningham, et al, 1993). We would assume that a constructivist learning environment in general is focused on higher order thinking related to originality and creativity. This higher order thinking has been described by Bloom (1956) in his taxonomy as the "synthesis" level, by Gagne (1987) as the "cognitive strategy" level and by Merrill (1983) as the "find" level. All of these models of higher order thinking stress the creation of new knowledge over lower level thinking skills such as recognition, paraphrasing, and application. Higher order thinking involving the construction and interpretation of knowledge, then, is the goal of a constructivist learning environment.
Hypermedia systems, although they generally bear the imprint (mental model) of their author, have the capability to move beyond an objectivist learning environment to a constructivist learning environment. One of the obvious strengths of hypermedia is that it is associative and may be designed to strongly reflect the semantic structure of a subject matter expert's knowledge. An equally obvious danger is that the semantic structure of the subject matter expert's biases or unawareness may also be reflected in the final application. Gender bias is a good example. Although hypermedia clearly provides the author with the capability to present multiple views in multiple modes, the medium itself is no assurance that the application will be developed well.
Because many educators are not technically oriented and have little if any application development experience before hypertext and hypermedia they may not focus enough on planning before starting to develop a hypermedia application. The importance of planning, whether for an entire automated system or for a single computer application, cannot be overstated. Computer system developers and application programmers learn early in their careers that successful systems and programs result from rigorous up front planning that fully determines the final output of the system and/or program. The hypertext/ hypermedia development paradigm which is characterised by multiple paths and outcomes through an application mandates an even higher emphasis on planning than most conventional applications. Some guidelines for constructing gender sensitive multimedia applications include:
Various forms of gender bias have been identified and are well known. The correction of gender bias in electronic text is now easier to accomplish through using tools such as the GENDER tool described in this paper. Incorporating the use of software tools in the review process for multimedia applications is now a logical and feasible step in the development process as evidenced by the capabilities and performance of the GENDER tool. By utilising this tool throughout the development process and following the guidelines for gender sensitive multimedia, developers can produce applications with far less gender bias than is typically found in most written texts and early computer, applications.
Proper planning is the single most important factor in the successful development of an interactive multimedia application. This planning should begin before the development process and question the specific educational environment and appropriateness of interactive multimedia for that situation. Planning should extend into the development process to insure that the various text and non-text modules are being assembled so they lead to the accomplishment of the goals identified for the project. Planning should also be included in the implementation stage to insure that there is a strategy for properly introducing students to the technical concepts and structure that underlie the content material.
Through careful analysis and planning at the beginning of an interactive multimedia project, course developers and educators can better meet the challenge of developing successful hypermedia projects. Interactive multimedia applications can flourish in educational environments if educators identify and then strive to meet the challenge of the seamless integration of information.
Bloom, B. S. (1956). Taxonomy of educational objectives. New York: Longman.
Britton, G. E. (1974). Why Jane can't win (Sex stereotyping and career role assignments in reading materials). Paper presented at the Annual Meeting of the International Reading Association, New Orleans, May 1.
Britton, G. & Lumpkin, M. (1976). For sale: Subliminal bias in textbooks. Reading Teacher, 31, 40-50.
Bundy, A. (1980). Artificial intelligence: An introductory course (pp. 59-120). Edinburgh: Edinburgh University Press.
Cunningham, D. S., Duffy, T. M. & Knuth, R. A. (1993). The textbook of the future. In McKnight, C., Dillon, A. & Richardson, J. (Eds), Hypertext: A psychological perspective. New York: Ellis Horwood.
Gagne, R. M. (1987). The conditions of learning (4th ed.) New York: Holt, Rinehart & Wilson.
Gonzalez-Suarez, M. & Ekstrom, R. B. (1989). Are US elementary school reading textbooks sex stereotyped? Presented at the Annual Meeting of the American Educational Research Association, San Francisco, March 27-3 1.
Gunderson, D. (1975). Reading and sex roles stereotypes. Paper presented at the Annual Meeting of the National Council of Teachers of English, San Diego, November 26.
Hughes, R. (1992). The fraying of America. Time, February 3, 44-49.
Jonassen, D. H. (1993). Effects of semantically structured hypertext knowledge bases. In McKnight, C., Dillon, A. & Richardson, J. (Eds), Hypertext: A psychological perspective. New York: Ellis Horwood.
Kepner, H. & Koehn, L. (1977). Sex roles in mathematics: A study of the status of sex stereotypes in elementary mathematics texts. Arithmetic Teacher, 24, 379-385.
Kernberger, C. (1990). Just say no to he/men, or teaching non-sexist language in the ESL classroom. Paper presented at the Annual Meeting of the New Mexico Teachers of English to Speakers of Other Languages, Las Cruces, February, 16-17.
Merrill, M. D. (1983). Component display theory. In Reigeluth, C. M. (Ed), Instructional design theories and models. Hillsdale, NJ: Lawrence Erlbaum Associates.
Pyle, W. J. (1976). Sexism in children's literature. Theory Into Practice, 15, 116-119.
Sadker, M., & Sadker, D. (1979). Beyond pictures and pronouns: Sexism in teacher education textbooks. Washington, DC: Women's Educational Equity Act Program, Office of Education, Department of Health, Education, and Welfare.
Sadker, M., Sadker, D., & Klein, S. (1991). The issue of gender in elementary and secondary education. Review of Research in Education, 17, 269-334.
Schau, C. G., & Scott, K. P. (1984). The impact of gender characteristics on instructional materials: An integration of the research literature. Journal of Educational Psychology, 76, 173-183.
Schutzer, D. (1987). Artificial intelligence: An applications oriented approach. New York: Van Nostrand Reinhold.
Scott, K. P. & Schau, C. G. (1985). Sex equity and sex bias instructional materials. In S. S. Klein (Ed), Handbook for achieving sex equity through education, 218-232. Baltimore: John Hopkins University Press.
Warren, C. R. (1988). The use of ordinary least squares regression in the development of an instrument to measure sexism in biology textbooks. Paper presented at the Annual Meeting of the Mid-West Educational Research Association, Chicago, October 13.
Weston, L. & Stein, S. (1978). A content analysis of publishers' guidelines for the elimination of sex role stereotyping. Educational Research, 7(3), 13-14.
|Authors: Gene McGuire, Instructor|
B. J. Gleason, Instructor
Computer Science and Information Systems Department,
The American University, 4400 Massachusetts Avenue,
NW Washington, DC 20016-8116 USA
Please cite as: McGuire, G. and Gleason, B. J. (1994). Using interactive multimedia to reduce gender bias in educational materials. In C. McBeath and R. Atkinson (Eds), Proceedings of the Second International Interactive Multimedia Symposium, 345-350. Perth, Western Australia, 23-28 January. Promaco Conventions. http://www.aset.org.au/confs/iims/1994/km/mcguire.html