# A Semi-Autoregressive Graph Generative Model for Dependency Graph Parsing

arXiv: https://arxiv.org/abs/2306.12018 · HTML: https://mlsanity.com/paper/2306.12018/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 41 references that are in the map.
- First posted on arXiv 21 Jun 2023.

## Builds on (most cited first, 25 of 25)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Semi-Supervised Classification with Graph Convolutional Networks](https://mlsanity.com/paper/1609.02907/) (2016)
- [Graph Attention Networks](https://mlsanity.com/paper/1710.10903/) (2017)
- [Unsupervised Cross-lingual Representation Learning at Scale](https://mlsanity.com/paper/1911.02116/) (2019)
- [Sequence Level Training with Recurrent Neural Networks](https://mlsanity.com/paper/1511.06732/) (2015)
- [Deep Biaffine Attention for Neural Dependency Parsing](https://mlsanity.com/paper/1611.01734/) (2016)
- [ReZero is All You Need: Fast Convergence at Large Depth](https://mlsanity.com/paper/2003.04887/) (2020)
- [75 Languages, 1 Model: Parsing Universal Dependencies Universally](https://mlsanity.com/paper/1904.02099/) (2019)
- [AMR Parsing as Sequence-to-Graph Transduction](https://mlsanity.com/paper/1905.08704/) (2019)
- [Efficient Second-Order TreeCRF for Neural Dependency Parsing](https://mlsanity.com/paper/2005.00975/) (2020)
- [AMR Parsing via Graph-Sequence Iterative Inference](https://mlsanity.com/paper/2004.05572/) (2020)
- [Stack-Pointer Networks for Dependency Parsing](https://mlsanity.com/paper/1805.01087/) (2018)
- [Order Matters: Probabilistic Modeling of Node Sequence for Graph Generation](https://mlsanity.com/paper/2106.06189/) (2021)
- [Broad-Coverage Semantic Parsing as Transduction](https://mlsanity.com/paper/1909.02607/) (2019)
- [Core Semantic First: A Top-down Approach for AMR Parsing](https://mlsanity.com/paper/1909.04303/) (2019)
- [Left-to-Right Dependency Parsing with Pointer Networks](https://mlsanity.com/paper/1903.08445/) (2019)
- [Second-Order Semantic Dependency Parsing with End-to-End Neural Networks](https://mlsanity.com/paper/1906.07880/) (2019)
- [Structured Training for Neural Network Transition-Based Parsing](https://mlsanity.com/paper/1506.06158/) (2015)
- [Global Greedy Dependency Parsing](https://mlsanity.com/paper/1911.08673/) (2019)
- [Global-aware Beam Search for Neural Abstractive Summarization](https://mlsanity.com/paper/2009.06891/) (2020)
- [Enhanced Universal Dependency Parsing with Second-Order Inference and Mixture of Training Data](https://mlsanity.com/paper/2006.01414/) (2020)
- [Easy-First Dependency Parsing with Hierarchical Tree LSTMs](https://mlsanity.com/paper/1603.00375/) (2016)
- [TGIF: Tree-Graph Integrated-Format Parser for Enhanced UD with Two-Stage Generic- to Individual-Language Finetuning](https://mlsanity.com/paper/2107.06907/) (2021)
- [Enhanced Universal Dependency Parsing with Automated Concatenation of Embeddings](https://mlsanity.com/paper/2107.02416/) (2021)
- [Bi-directional Attention with Agreement for Dependency Parsing](https://mlsanity.com/paper/1608.02076/) (2016)

## Abstract (arXiv)

Recent years have witnessed the impressive progress in Neural Dependency Parsing. According to the different factorization approaches to the graph joint probabilities, existing parsers can be roughly divided into autoregressive and non-autoregressive patterns. The former means that the graph should be factorized into multiple sequentially dependent components, then it can be built up component by component. And the latter assumes these components to be independent so that they can be outputted in a one-shot manner. However, when treating the directed edge as an explicit dependency relationship, we discover that there is a mixture of independent and interdependent components in the dependency graph, signifying that both aforementioned models fail to precisely capture the explicit dependencies among nodes and edges. Based on this property, we design a Semi-Autoregressive Dependency Parser to generate dependency graphs via adding node groups and edge groups autoregressively while pouring out all group elements in parallel. The model gains a trade-off between non-autoregression and autoregression, which respectively suffer from the lack of target inter-dependencies and the uncertainty of graph generation orders. The experiments show the proposed parser outperforms strong baselines on Enhanced Universal Dependencies of multiple languages, especially achieving 4% average promotion at graph-level accuracy. Also, the performances of model variations show the importance of specific parts.