Abstract
The 2018 NAACL-HLT conference represented a pivotal moment for the Natural Language Processing (NLP) community, characterized by unprecedented growth and a strategic shift toward industrial applications and ethical considerations. This volume of short papers captures the state of the field during the emergence of contextualized word representations and the formalization of reproducibility and diversity initiatives within the conference structure.
Introduction
NAACL-HLT 2018, held in New Orleans, was the largest iteration of the conference to date. With 1,122 research track submissions, the event signaled a "breaking point" that necessitated structural changes to handle the explosive growth of the field. This volume reflects a community grappling with its own maturity, balancing academic rigor with the burgeoning needs of the industrial sector.
Organizational Innovations and the Industrial Track
To address the scalability crisis, the program committee, led by General Chair Marilyn Walker and Program Co-Chairs Heng Ji and Amanda Stent, introduced several key innovations. A primary addition was the dedicated Industrial Track, designed for scalable, reliable, and customer-facing applications. This move acknowledged the deep integration of NLP research into commercial products.
Beyond the industrial focus, the conference implemented structured reviewing to improve feedback quality, requiring authors to address software and data handling. Reproducibility was further championed by encouraging code uploads. Diversity initiatives were also prioritized, including family-friendly policies like childcare and subsidies for researchers from developing countries to ensure broader participation.
Research Highlights: The ELMo Moment and Ethical Pivots
A defining moment of the conference was the recognition of "Deep Contextualized Word Representations" (Peters et al.), commonly known as ELMo. This work shifted the paradigm from static embeddings to context-dependent representations, laying the groundwork for the Transformer-based models that followed.
Simultaneously, the program demonstrated an increased focus on AI ethics. Early papers in the volume addressed gender bias in coreference resolution, signaling that fairness had become a core technical challenge. The conference also showcased the expansion of NLP into clinical domains, such as detecting Alzheimer’s dementia through neuropsychological exam analysis.
The community also looked backward to honor foundational work through "Test of Time" awards. These recognized papers from 2002–2012 that continue to shape the industry, such as the introduction of the BLEU metric for machine translation and early sentiment analysis techniques.
Datasets and Linguistic Diversity
The proceedings introduced several specialized datasets, moving beyond standard benchmarks. Notable examples include corpora for biological laboratory protocols, metaphorical language in non-native English, and semantic similarity in Vietnamese. These efforts reflect a push toward diversifying the tasks and languages addressed by the community, though the organizers noted that the field remains heavily English-centric.
Conclusion
While NAACL-HLT 2018 celebrated significant technical and organizational achievements, it also highlighted ongoing challenges. The peer-review infrastructure remains under strain, requiring a workforce of over 1,300 reviewers. Future efforts must leverage NLP's own tools—such as text mining and network analysis—to optimize reviewer matching and continue fostering global linguistic diversity.
Summary updated