📄 Investigating the Use of English Prepositions by Iraqi IT Students Learning English as a Foreign Language

Summary

The study asks how well Iraqi university students in an IT programme handle English prepositions. Arabic-speaking EFL learners often pick the wrong one because preposition use depends on context and differs from their native language. The author first reviews the grammar literature: definitions, the relations prepositions express, and why they are hard. Reasons include polysemy, metaphorical extension from a core sense, short unstressed forms that are easy to miss in speech, and negative transfer from the mother tongue. The empirical part is a multiple-choice test given to 40 randomly chosen third-year Software Engineering students at the University of Nineveh. It has 15 gap-fill sentences, evenly split across prepositions of time, place and "connection", each with four options. Every category scored well below half correct, and connection-type prepositions were the weakest. Both hypotheses were confirmed. The author attributes the errors mainly to Arabic–English differences and recommends that syllabi and teachers give prepositions explicit, sustained attention with varied practice.

Key points

• The sample was 40 of 50 third-year students in the Department of Software Engineering, College of Information Technology, University of Nineveh (Mosul).
• The instrument had 15 MCQ items (5 time, 5 place, 5 connection), each with four options. It was checked by a jury and administered in a 45-minute supervised session.
• Mean correct rates were 36% for time, 29% for place and 26% for connection, giving error rates of 64%, 71% and 74%.
• Item difficulty varied widely:
  • "careful in the future" had 70% correct (28/40).
  • "return the book by April 21st" had 2.5% correct (1/40).
  • "a boy of 15" had only 10% correct.
• The literature review draws on Martinet's two-step difficulty: first deciding whether a preposition is needed, then choosing which one. It also uses Lakoff's view that senses extend metaphorically from a prototypical meaning.

Interesting ideas

• I1. The same form (after, before, since, until, as) is a preposition when a noun phrase follows and a subordinator when a clause with subject and verb follows, so the right-hand context decides the category.
• I2. Multiword prepositions such as "instead of", "next to" or "in front of" function as one relational unit, and many verb+preposition combinations are fixed idioms rather than freely assembled pairs.
• I3. Prepositions have a small core of literal senses, and other senses are metaphorical extensions of the prototype, such as spatial "in" or "at" reused for time.
• I4. Splitting preposition choice into two decisions, whether a preposition is needed and then which one, isolates two different error types.
• I5. A small balanced cloze probe with five items per semantic class and four distractors, scored per item and per class, quickly reveals which relation types are weakest. Per-item spread (2.5% to 70%) was more informative than class averages.

For Mova

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

• I1 (tokens): The tagger can use a cheap feature on whether the next material contains a finite verb with its subject, which helps choose ADP vs SCONJ for these words and gives an explainable reason for the tag.
• I2 (tree): Keep a lexicon of multiword prepositions and verb–preposition idioms to drive consistent fixed and oblique attachments when converting treebank dialects and checking parser output.
• I3 (world): Store a prototype sense per preposition with explicit extension links (place→time, place→abstract state), so that entity and event placement in the text world can say which extension it applied.
• I4 (back): Generation from the meaning graph can first decide whether the relation surfaces as a bare argument or a prepositional phrase, then pick the lexical item, logging each decision separately for debugging.
• I5 (learn): Build a similar per-class preposition gap-fill probe as a regression check for the tagger and parser, and report item-level accuracy so that rare usages like deadline "by" or age "of" are not hidden by averages.