Search by Fuzzy Inference in a Children's Dictionary

Abstract

This research aims at promoting the usage of an online children's dictionary within a context of reading comprehension and vocabulary acquisition. Inspired by document retrieval approaches developed in the area of information retrieval (IR) research, we adapt a particular IR strategy, based on fuzzy logic, to a search in the electronic dictionary. From an unknown word, searched for by a learner, our proposed fuzzy inference process makes it possible to retrieve relevant lexical information from any entry of the dictionary. Furthermore, it organises this information in the form of semantic maps of an adaptable size surrounding the query word. Manual construction of such semantic maps are seen as being effective for helping learners in vocabulary acquisition and reading comprehension tasks. Our research leads to a capability for building them automatically. Using concrete examples, we provide details of the calculation for the construction of semantic maps as well as for the retrieval of information. We introduce a software component which could be integrated in a computer assisted language learning (CALL) environment to promote vocabulary acquisition in L1.

Notes

See www.tea.state.tx.us/student.assessment/resources/online/2003/grade3/read.htm.

Copyright obtained from Houghton Mifflin for an electronic reproduction of the American Heritage First Dictionary, originally in paper form, to be used for research purposes.

Hayes and Ahrens (1988) they identify 30.9 rare words per 1,000 in a children's book. In comparison, newspapers have 68.3, and prime-time adult shows 22.7.

In the Edinburgh Association Thesaurus (www.eat.rl.ac.uk), empirical data is gathered about word associations. When egg is the trigger word, top associations are with yolk, hen, cup, bacon and chicken.

You can find this story at www.magickeys.com/books/farm/index.html.

In the example, we voluntary omit the significant words “straight”, “usually”, “Tom” and “Jim” from Table 1 only to limit the size of the table. Membership degree to a fuzzy set is always defined by the interval of values μ _e (t) →[0,1]. So, we should normalise the frequencies of occurrences by a large positive number N, but according to Miyamoto, this parameter disappears in the calculation of the fuzzy similarity. For the mathematical details of the automatic generation of a pseudo-thesaurus, we refer the reader to Miyamoto (1990).

We refer the reader to Klir and Yuan (1995) for the mathematical explanation of the composition of fuzzy relations.

We give Boolean values to the terms in the original query; otherwise we would have to find a way to assign a weight to them with respect to the user's expectation or certainty. This is outside the scope of the type of search assumed here by a young reader in a dictionary

In the program, we normalise the weight of terms by the highest frequency of a word found across every vector of the AHFD. The max – min composition can be used as the operator of what St-Jacques and Barrière (2004) called the “dictionaric inference” (Ax R By) when we infer from a dictionary A(t_j )°B(t_i,t_j ). To be precise, we should say that the α-cut is applied to the values of P(t_i ,t_j ) before obtaining A by max – min composition: A = Q°P except that the min value is always from the pseudo-thesaurus because we give a Boolean value (1) to the query's words. By extension, we are saying here that α-cut is also the limitation for the weight of the terms in an expanded query A_α .