Using corpora in teaching language has been shown to be effective for already known items such as words and collocations. An exploratory study by Kuei-Ju Tsai published in 2019 looked at how well students learn new vocabulary using a corpus and a dictionary.

Article

Kuei-Ju Tsai (2019): Corpora and dictionaries as learning aids: inductive versus deductive approaches to constructing vocabulary knowledge, Computer Assisted Language Learning, DOI: 10.1080/09588221.2018.1527366

Introduction

Previous work on DDL (data driven learning) has shown that using concordances is good for consolidating already known language knowledge. For example a set of concordances of the word “proximity” would show the association to “in close”. However if a student does not initially know what “proximity” means using a concordance to work out its meaning would not be straightforward, a dictionary would be much faster.

Using concordances to develop vocabulary knowledge in depth is said to use an inductive approach – particular and multiple language items are examined to infer generalisations. By contrast using a dictionary is a deductive approach – the generalisation is given immediately, then users see if it applies to their particular case.

The inductive approach involves more work on the part of the learner (compared to being given “answers” using a deductive approach) and hence is thought to lead to better learning effects.

Although both inductive and deductive approaches have been shown to be effective for receptive vocabulary knowledge only inductive approaches have more evidence for improving productive vocabulary knowledge especially for collocations and colligations (lexico-grammatical knowledge).

Study

• Participants: Two intact classes of 100 students from a university English class in Taiwan for the quantitative aspect of study. Plus 12 different students for the qualitative aspect of study. All students at A2 CEFR level (TOEIC scores of 550 or 590).

• Instruments: From an initial list of 100 words from BNC-COCA 6k-level wordlist a short list of 5 words (confirmed to be unknown to all the participants) finally selected – accrue, acquit, dementia, proximity, stringent. Post-test – 6 tests on word knowledge – word form recall, part-of-speech (POS) recognition, definition recall, definition recognition, collocation recall, and collocation recognition (in this order)

• Treatment: Students were presented with a word for 10 mins each and asked to learn them either using first the COCA corpus and then the Macmillan online dictionary (inductive group) or first the dictionary and then the corpus (deductive group). The inductive group were asked to infer meaning from the corpus search and then verify the meaning using the dictionary, the deductive group looked up the meaning of a word first in the dictionary and then asked to look for 3 concordance examples in COCA which best illustrated the meaning. The 12 students in the qualitative part did a think-aloud protocol as they either did the inductive or deductive treatment. The think-aloud group screen activities of corpus/dictionary queries, search results, and verbal reports were captured by a screen recorder. All the students saw a tutorial on how to use COCA and then did a random search to try out the corpus tool. The think-aloud group also got a video tutorial for how to do the think-aloud protocol. A surprise post-test was given 2 weeks later.

• Measures collected: A correct answer in the test got 1 point, an incorrect answer got no points. Each test had a maximum of 5 points, with the six tests that made a maximum of 30 points available. The think-aloud reports were coded initially according to Paul Nation’s word knowledge framework and then refined as other themes in the learners actions emerged. Inter-rater reliability from 3 raters was 0.826. The coding scheme was as follows: Browsing concordances; Browsing the extended contexts of a particular concordance 
line (Context+); Browsing collocations; Translating; Inferring POS; Inferring word associations; Inferring word definitions. The last three categories were not used for the deductive group as they did not need to do any inferring.

• Results: The two groups did not differ (statistically) significantly on overall score from the 6 tests. They did differ significantly on 2 tests – the deductive group did better on definition recall and the inductive group did better on collocation recall. For the six participants recording their thoughts when using the corpus there was considerable variation – one student looked at 123 concordance lines and translated 31 instances while another only looked at 38 lines and translated twice.

Overall the results suggest that the inductive approach is better for learning collocations and the deductive approach better for building definitional knowledge. Previous research confirms this broad conclusion.

So what?

From the fact that different approaches, inductive and deductive, are good for different purposes one could reason that a tool that included opportunities for both would be useful. A previous exploratory study concluded a new generation of tools that integrated multiple sources of information in a useable package needs developing.

An example of a tool that is available now that contains elements that can be used for both deductive and inductive learning is the iWeb corpus, from the same people who brought you the COCA corpus. This online resource allows a student to access definitions, pronunciations, synonyms, collocates, topics, clusters, websites, and concordance lines either on one page or separate windows for each feature (see screenshot below):

There are some issues with this particular tool as noted by respected lexicographer . The biggest is that the definitions used by iWeb (from vocabulary.com) are not as good for learners as established sources such as the Macmillan dictionary. In addition the interface of iWeb could be much cleaner, as it is, it is very busy and might put off some learners.