As the demand for publishing in English grows across STEM fields, researchers who use English as a second language face persistent writing challenges, including limited mastery of domain-specific terminology, a tendency toward literal translation, and sentence-length and error issues originating from compact rhetorical habits in their native language. Recent advances in artificial intelligence and data-driven methods offer new opportunities for language instruction and writing support. This study presents a comprehensive review of traditional corpus-building approaches and recent data-driven AI techniques, and then provides a focused case discussion in marine engineering conducted in collaboration with domain researchers. By comparing corpus resources, feature engineering strategies, and modeling approaches, we evaluate the transferability of data-driven instructional models for improving publication readiness among non-native engineering researchers. The discussion highlights evidence-based revision strategies, targeted pathways for disciplinary vocabulary acquisition, and practical ways to integrate automated feedback with expert human guidance to improve manuscript quality. Our review indicates that data-driven AI tools hold promise for supporting terminology acquisition, identifying common error patterns, and generating actionable revision suggestions, but their effectiveness depends on high-quality, domain-specific corpora and close collaboration with subject-matter experts. We conclude with recommendations for future research and pedagogical practice to scale evidence-based ERPP (English for Research Publication Purposes) support across STEM disciplines.