Overview of Meituan's ACL 2021 Accepted Papers

ACL (Association for Computational Linguistics) is the top international conference in computational linguistics and natural language processing (NLP). According to Google Scholar metrics, ACL ranks first in impact and is a CCF‑A recommended conference. In 2021, Meituan's technology team had seven papers (six long papers and one short paper) accepted at ACL, covering event extraction, entity recognition, intent detection, novel slot discovery, unsupervised sentence representation, semantic parsing, and document retrieval.

Accepted papers:

Capturing Event Argument Interaction via A Bi‑Directional Entity‑Level Recurrent Decoder (Long Oral)

Slot Transferability for Cross‑domain Slot Filling (Findings Long)

Modeling Discriminative Representations for Out‑of‑Domain Detection with Supervised Contrastive Learning (Short Poster)

Novel Slot Detection: A Benchmark for Discovering Unknown Slot Types in the Task‑Oriented Dialogue System (Long Oral)

ConSERT: A Contrastive Framework for Self‑Supervised Sentence Representation Transfer (Long Poster)

From Paraphrasing to Semantic Parsing: Unsupervised Semantic Parsing via Synchronous Semantic Decoding (Long)

Improving Document Representations by Generating Pseudo Query Embeddings for Dense Retrieval (Long Oral)

Each paper introduces a novel method: the first proposes a bi‑directional entity‑level decoder (BERD) for event argument generation; the second defines “slot transferability” to select source‑task slots that best transfer to a target task; the third applies supervised contrastive learning to improve unsupervised out‑of‑domain intent detection; the fourth defines the Novel Slot Detection (NSD) task and provides benchmark datasets; the fifth presents ConSERT, a contrastive self‑supervised approach that mitigates BERT’s sentence‑embedding collapse; the sixth introduces Synchronous Semantic Decoding (SSD) to perform unsupervised semantic parsing by treating it as a constrained paraphrasing problem; the seventh generates multiple pseudo‑query embeddings for each document to enhance dense retrieval efficiency and effectiveness.

The article concludes by encouraging researchers interested in these topics to read the detailed papers and invites further collaboration with Meituan’s research team (contact: [email protected]).