📄 Learning English Grammar by Means of M-Learning: A Case Study

Summary

This case study asks whether mobile apps help Ukrainian students at a non-linguistic university learn English grammar better than paper-based teaching. At Sumy National Agrarian University, 71 second-year students aged 18–19 took part. Two experimental groups (35 students) used tablets and phones with Wi-Fi for two semesters. Their apps included British Council's Learn English Grammar and Johnny Grammar's Word Challenge, English Grammar in Use, Mad Libs, and messengers for handing out tasks and giving feedback. Two control groups (36 students) worked with textbooks, paper dictionaries and teacher materials. A pre-test showed the groups started at similar levels. Final results came from tests, oral answers and control works. Each student received a knowledge ratio averaging three sub-scores, which was mapped to four proficiency bands. Groups were compared with a percentage-difference t statistic. The experimental groups ended mostly at sufficient or high levels, while the control groups stayed mostly low or initial. Interviews showed students largely liked the apps but named some drawbacks. The authors argue that m-learning works for grammar, but teachers need methodological preparation to use it well.

Key points

• The sample had 71 volunteers: 35 in experimental groups and 36 in control groups. All were second-year students in 2021–2022 with 2 EFL classes a week on the same syllabus.
• The knowledge ratio is RK = (a + b + c)/3, where the parts are systematics, volume and efficacy. Score bands were: high if 3.67 < RK ≤ 5, sufficient if 2.33 < RK ≤ 3.67, low if 1 < RK ≤ 2.33, and initial otherwise.
• Group differences were judged with fixed thresholds: t > 2 counts as significant and t > 3 as definitely reliable.
• Experimental group results were 24% high, 46% sufficient, 25% low and 5% initial.
• In the survey:
  • 96.1% valued combining study with gadgets.
  • 92.7% reported higher motivation.
  • 96% wanted to keep using the apps.
  • 79.3% valued chat-bot analysis of their mistakes.
  • 81.2% reported distraction by entertainment apps.
  • 69.3% complained about small screens.

Interesting ideas

• I1. Score each learner on three separate axes (systematics, volume, efficacy), average them, and map the result to named bands; one number then reflects breadth, organisation and practical success together.
• I2. Run a pre-test to confirm that the compared groups start out equivalent before any intervention, so a later gap can be credited to the treatment.
• I3. Compare two percentages with a simple t statistic and fixed thresholds (2 for significant, 3 for reliable) rather than reading raw differences by eye.
• I4. Split grammar into an active minimum that students must produce and a larger passive minimum they only need to understand.
• I5. The Mad Libs game has one player supply words by category to fill story gaps, guided by hints. It is a playful cloze task that tests grammatical fit.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (answer): School-course readers could be graded on separate axes such as coverage, consistency of reasoning and correctness of answers, then reported as levels instead of a single accuracy figure.
• I2 (learn): Before comparing two parser or training variants, check that both start from the same baseline and data split, which works like a negative control on the setup.
• I3 (learn): LAS or QA-accuracy gains between model versions could go through such a cheap significance gate before a change is accepted.
• I4 (back): The generator could limit itself to a small, safe set of constructions, while the parser and tree stage are still required to handle the much wider passive range.
• I5 (frames): Mad Libs-style probes could blank out frame slots in a fable and check whether the role constraints from VerbNet accept or reject the candidate fillers.