📄 Positive Evidence and Parameter Resetting in the Teaching of English Grammar to Libyan University Students

Summary

This study tests whether grammar teaching that shows learners only correct sentences, with no marked ungrammatical examples, lets adult Arabic speakers acquire English grammatical settings that differ from Arabic. It uses Chomsky's principles-and-parameters framework and focuses on two parameters. The first is pro-drop: Arabic allows null subjects, while English requires overt ones. The second is verb raising: English main verbs cannot move past adverbs, so the order subject–verb–adverb–object is ruled out.

Twenty randomly selected Advanced Grammar students at Misurata University, Libya, completed two written grammaticality judgment tasks, one per parameter. The authors expected parameter resetting to fail under positive-evidence-only instruction. They also expected that one parameter might prove easier than the other.

Both parameters appeared largely unreset. The pro-drop task went slightly better, which the authors explain by the visibility of the obligatory English subject. Correct input alone never signals that an adverb cannot sit between verb and object. The authors therefore recommend adding negative evidence and grammaticality judgment tests to grammar teaching.

Key points

• Participants were 20 Arabic-L1 advanced students of English at Misurata University. Each completed two grammaticality judgment (GJ) tasks, one for pro-drop and one for verb raising.
• On the verb-raising task, per-student scores were low:
  • 2 students scored 0%
  • 11 students scored 16%
  • 3 students scored 33%
  • 3 students scored 50%
  • only 1 student reached 66.67%
• Item results for the verb-raising task:
  • The best-rejected item, with an adverb between verb and object, was still turned down by only 45% (9 of 20).
  • Other items were rejected by 30–35%.
  • The sentence with "explains slowly the lesson" was rejected by just 10%.
• No participant could explain why they rejected any sentence.
• Most students performed similarly on the two tasks, with pro-drop somewhat better. The authors conclude that neither parameter was convincingly reset.

Interesting ideas

• I1. Seeing only well-formed examples cannot reveal which orders are forbidden, so learning needs explicit negative examples. — For Mova (learn): Training and evaluation sets for the parser and repair rules should include deliberately ungrammatical minimal variants as negative controls, so we can check that the system learned constraints and not just frequencies.
• I2. Some constraints are salient in positive data, like the obligatory subject, and others are nearly invisible, like the ban on an adverb before the object. Learnability depends on how visible the contrast is in the input. — For Mova (learn): We could estimate per construction how detectable its constraint is from silver data, and target extra teacher or negative data at the invisible ones.
• I3. The ban on placing a manner or frequency adverb between a main verb and its direct object is a crisp, testable rule. — For Mova (back): The generator and compressor output could be checked against this rule, so regenerated text never yields orders like "ate quickly a pear".
• I4. Learner-like SVAO sentences are realistic ill-formed input that readers still understand. — For Mova (tree): Robustness tests could check that parsers still attach the displaced adverb to the verb and keep the object relation, instead of misparsing the whole clause.
• I5. Judgments with no stated reason were treated as a sign of shallow competence. — For Mova (answer): When a student module accepts or rejects a sentence, it should output the rule that justifies the verdict, making an unexplained correct guess distinguishable from real knowledge.