For decades, test developers have attempted to create a generalized electronic scoring system for texts. It's even been called the "holy grail" of educational assessment because it would free us from the tyranny of multiple-choice items. Products developed exclusively with data analytics, like IBM's Merlin, are making some inroads, but they require huge amounts of text to produce results and aren't yet reliable enough for use in individual assessment. Moreover, like the electronic scoring systems now being produced by educational test developers, they measure easily quantified aspects of texts (like vocabulary and punctuation) rather than what we really want to know—how deeply students are learning the skills and knowledge they need for success in life and work.
CLAS is the first computerized developmental scoring system that measures what educators really want to know—how well learners can put their knowledge to work in real-world contexts—reliably enough to be used ethically in individual assessment contexts. It's a truly general system that can readily be adapted for use in diverse knowledge domains. Rather than measuring whatever aspects of texts are most easily quantified, CLAS measures growth along a well-validated developmental scale called the Lectical Scale. Its algorithms are based on a deep study of learning and development that involves a blend of human expertise and analytics. When CLAS produces a score, it tells us where an assessment performance lands on the Lectical Scale, what the score means in terms of the development of a test-taker's skills, their understanding of concepts targeted by the assessment, and what the test taker is likely to benefit from learning next.
CLAS algorithms are based on human and computer analysis of tens of thousands of texts (and growing). At its core is the Lectical™ Dictionary—the world's first developmental dictionary—a constantly evolving taxonomy of the development of meanings. (More about the Dictionary below.)
Like the LAS (Lectical Assessment System, our human scoring system), CLAS can be used:
CLAS has several advantages over the LAS:
Since its introduction in 2014, CLAS has been used to score thousands of assessments for several research and evaluation projects.
The Lectical Dictionary is at the core of everything we do here at Lectica. It not only makes CLAS possible, but we use it to describe learning sequences and develop customized learning resources and other report feedback. We've also used it to create a better spell checker—one that takes the developmental level of performance into account when it makes spelling suggestions.
The Dictionary is composed of units of meaning called "Lectical Items"—words or phrases like "evidence" and "reliable evidence" that carry meaning. Each Lectical Item is assigned to a Lectical Phase (1/4 of a Lectical Level), based on a combination of empirical evidence, the judgment of our analysts, and a variety of helper algorithms. The goal is to assign items to the lowest level at which the simplest meaning they carry is likely to be useful.
For example, an examination of items containing the word evidence reveals easily observed progressions in the development of its meaning, such as the following:
Lectical Dictionary entries begin with first speech and cover the full span of development. In addition to being assigned to a Lectical Phase, many of the items in the Lectical Dictionary have been assigned to thematic strands such as deliberation, conflict resolution, the physics of energy, or evidence. For example, at this writing, there are over 85,000 terms in the Lectical Dictionary that relate to evidence. Longitudinal and cross-sectional analyses of sequences like the evidence sequence has demonstrated that each successive conception builds upon previous conceptions (Dawson & Gabrielian, 2003; Dawson- Tunik, 2004). These findings are consistent with the developmental theory upon which the dictionary is based (Fischer, 1980; Piaget, 1985), and suggest that Lectical Items assigned to a particular phase can be said not only to represent the understandings of that phase but also play the role of building blocks for future conceptions.
The distribution of Lectical Items from different phases within a given performance can, therefore, be thought of as evidence of the historical pattern of an individual’s development. The rate of dictionary development has increased as what we learn about patterns in the acquisition of concepts is gradually integrated into our methods and technology. For example, we have learned that there are regularities in the progression of verb conjugation within particular developmental levels and that, in some cases, new single-word Lectical Items typically don’t appear next to “and” or “or” until the phase following the phase to which they have been assigned. Patterns like these, when adequately regular, allow a degree of automation in the lexicating process. The Lectical Dictionary is continuously monitored, refined, and added to by a team of trained analysts. Over time, as we build new assessments in new subject areas, the Lectical Dictionary will become an increasingly comprehensive curated taxonomy of learning.
As of May 2021, the Dictionary contained over a million curated items, which CLAS uses to score 6 different assessments.
Unlike the purely computational scoring algorithms of conventional "big data" analysis or "meta-analytics," CLAS's algorithms are developed through a human/machine collaboration. CLAS and our analysts are constantly engaged in "conversations" about the placement of Lectical Items. CLAS's algorithms represent our current knowledge, and our analysts interact with CLAS to integrate new knowledge. This partnership approach required a great deal of human labor and learning upfront, but the resulting system produces scores that are more accurate and much more meaningful than those produced with analytics alone.
In addition to providing a Lectical Score, CLAS provides several indicators of confidence in that score. We use these indicators to help us determine when a score needs human review and to identify cheating. The application that computes confidence statistics documents word count, calculates the density of spelling errors, and provides several statistics that allow us to determine if the curve for a given performance deviates from an expected pattern.
CLAS is not only making it possible to fulfill our educational mission—to optimize learning for everyone—but its technology will be useful in a wider range of contexts, including marketing, software, and psychological and sociological research. Over the next few years, we'll be exploring a range of applications.
If you'd like to know more about CLAS or have ideas about how CLAS technology could enhance your work, please contact us.
IES (US Department of Education)
The Spencer Foundation
NIH
Dr. Sharon Solloway
The Simpson Foundation
The Leopold Foundation
Glastonbury School District, CT
The Ross School
Rainbow Community School
The Study School
Long Trail School
The US Naval Academy
The City of Edmonton, Alberta
The US Federal Government
Antonio Battro, MD, Ph.D., One Laptop Per Child
Marc Schwartz, Ph.D. and former high school teacher, University of Texas at Arlington
Mary Helen Immordino-Yang, Ed.D., University of Southern California
Willis Overton, Ph.D., Temple University, Emeritus