Word Maturity: A New Metric for Word Knowledge

Abstract

A new metric, Word Maturity, estimates the development by individual students of knowledge of every word in a large corpus. The metric is constructed by Latent Semantic Analysis modeling of word knowledge as a function of the reading that a simulated learner has done and is calibrated by its developing closeness in information content to that of a simulated literate adult. Individual human learner knowledge is aligned with the simulation by adaptive testing. Evidence of accuracy, example applications to vocabulary assessment, teaching and reading research, properties of the metric, and a conjecture about its possible wider importance are described.This article introduces a new metric for the development of readingcomprehension-oriented word knowledge. In brief, a computational simulation based on Latent Semantic Analysis (LSA) first creates learning trajectories for each unique orthographic word-form in a text corpus that is representative in size and content of the lifetime reading of a literate adult. A metric-based adaptive test is used to align individual reader knowledge with the simulation. The technology can then make estimates of degree of knowledge of every individual word for individual readers relative to that of an average literate adult.We begin with a brief discussion of the need and value of such a metric, comparing it with the current state of the art in vocabulary assessment. Next come sections on LSA and how it is used to create the new metric. This includes a review of evidence that LSA does the things that the metric requires. This is Correspondence should be sent to Thomas K. Landauer, Pearson Knowledge Technologies, 4940 Pearl East Circle, Suite 200, Boulder, CO 80301.

Word maturity indices with latent semantic analysis: why, when, and where is Procrustes rotation applied?

“…This is a technical point of Procrustes alignment, which makes it necessary to have the same dimensions for both spaces. This could be justified by the theoretical assumption that there is an fixed optimal number of about 300 dimensions for general knowledge spaces 37 and even for the good predictions done with studies carried on with this constrain, 3,9,18,19 but this question and its potential consequences remain open, and may be answered in the future by applying a rotation technique without this restriction. For example, another form of alignment might be the 'conceptual alignment' which could be carried out using the technique suggested by a recent study.…”

“…The following criterion can be used to regard a word as having reached the mature meaning: a term is mature when its maturity curve surpasses .65, 3,19 that is, when it has acquired 65% of its meaning ( Figure 11, the curve for the term 'bosque' ('forest') reaches .65 at the age of 6). This is the age which is called time to maturity (TTM) in the model.…”

“…It was proposed by Kireyev and Landauer 18 and was recently expanded in other studies. 3,9,19,20 WM is based on the representation of word meaning provided by the latent semantic analysis (LSA) models. [21][22][23] However, apart from the LSA model itself, this methodology was not sufficiently widespread among the scientific community through these previous studies.…”

Predicting Word Maturity from Frequency and Semantic Diversity: A Computational Study

“…In other words, TTM corresponds to the age in which a particular word is used in an adult-like manner. TTM values have also provided good evidence of criterion validity in exhaustive studies (Biemiller et al, 2014;Landauer et al, 2011) using the Age of Acquisition (AoA) of words as the external criterion. Maturity curves based on the WM point for four words from our study.…”

“…But to do that, we had to previously replicate the word maturity paradigm (Biemiller et al, 2014;Landauer et al, 2011) using our own incremental spaces, this time with Spanish text samples. To this end, we followed the entire process for extracting the three Word Maturation indices (WM) for each term.…”