More Like this

1. Do Attention Heads in BERT Track Syntactic Dependencies?

   no surname), Phu Mon; Phang, Jason; Bordia, Shikha; Bowman, Samuel R. (January 2019, NY Academy of Sciences NLP, Dialog, and Speech Workshop)

   null (Ed.)

   We investigate the extent to which individual attention heads in pretrained transformer language models, such as BERT and RoBERTa, implicitly capture syntactic dependency relations. We employ two methods---taking the maximum attention weight and computing the maximum spanning tree---to extract implicit dependency relations from the attention weights of each layer/head, and compare them to the ground-truth Universal Dependency (UD) trees. We show that, for some UD relation types, there exist heads that can recover the dependency type significantly better than baselines on parsed English text, suggesting that some self-attention heads act as a proxy for syntactic structure. We also analyze BERT fine-tuned on two datasets---the syntax-oriented CoLA and the semantics-oriented MNLI---to investigate whether fine-tuning affects the patterns of their self-attention, but we do not observe substantial differences in the overall dependency relations extracted using our methods. Our results suggest that these models have some specialist attention heads that track individual dependency types, but no generalist head that performs holistic parsing significantly better than a trivial baseline, and that analyzing attention weights directly may not reveal much of the syntactic knowledge that BERT-style models are known to learn. 

   more » « less
2. Moment Distributionally Robust Tree Structured Prediction

   Li, Yeshu; Saeed, Danyal; Zhang, Xinhua; Ziebart, Brian D.; Gimpel, Kevin (December 2022, Advances in neural information processing systems)

   Structured prediction of tree-shaped objects is heavily studied under the name of syntactic dependency parsing. Current practice based on maximum likelihood or margin is either agnostic to or inconsistent with the evaluation loss. Risk minimization alleviates the discrepancy between training and test objectives but typically induces a non-convex problem. These approaches adopt explicit regularization to combat overfitting without probabilistic interpretation. We propose a momentbased distributionally robust optimization approach for tree structured prediction, where the worst-case expected loss over a set of distributions within bounded moment divergence from the empirical distribution is minimized. We develop efficient algorithms for arborescences and other variants of trees. We derive Fisher consistency, convergence rates and generalization bounds for our proposed method. We evaluate its empirical effectiveness on dependency parsing benchmarks. 

   more » « less
3. Discourse Analysis via Questions and Answers: Parsing Dependency Structures of Questions Under Discussion

   Ko, Wei-Jen; Wu, Yating; Dalton, Cutter; Srinivas, Dananjay; Durrett, Greg; Li, Junyi Jessy (July 2023, Findings of the Association for Computational Linguistics: ACL 2023)

   Automatic discourse processing is bottlenecked by data: current discourse formalisms pose highly demanding annotation tasks involving large taxonomies of discourse relations, making them inaccessible to lay annotators. This work instead adopts the linguistic framework of Questions Under Discussion (QUD) for discourse analysis and seeks to derive QUD structures automatically. QUD views each sentence as an answer to a question triggered in prior context; thus, we characterize relationships between sentences as free-form questions, in contrast to exhaustive fine-grained taxonomies. We develop the first-of-its-kind QUD parser that derives a dependency structure of questions over full documents, trained using a large, crowdsourced question-answering dataset DCQA (Ko et al., 2022). Human evaluation results show that QUD dependency parsing is possible for language models trained with this crowdsourced, generalizable annotation scheme. We illustrate how our QUD structure is distinct from RST trees, and demonstrate the utility of QUD analysis in the context of document simplification. Our findings show that QUD parsing is an appealing alternative for automatic discourse processing. 

   more » « less
4. Lightweight multi-language syntax transformation with parser parser combinators

   https://doi.org/10.1145/3314221.3314589  

   van Tonder, Rijnard; Le Goues, Claire (July 2019, Proceedings of the 40th ACM SIGPLAN Conference on Programming Language Design and Implementation)

   Automatically transforming programs is hard, yet critical for automated program refactoring, rewriting, and repair. Multi-language syntax transformation is especially hard due to heterogeneous representations in syntax, parse trees, and abstract syntax trees (ASTs). Our insight is that the problem can be decomposed such that (1) a common grammar expresses the central context-free language (CFL) properties shared by many contemporary languages and (2) open extension points in the grammar allow customizing syntax (e.g., for balanced delimiters) and hooks in smaller parsers to handle language-specific syntax (e.g., for comments). Our key contribution operationalizes this decomposition using a Parser Parser combinator (PPC), a mechanism that generates parsers for matching syntactic fragments in source code by parsing declarative user-supplied templates. This allows our approach to detach from translating input programs to any particular abstract syntax tree representation, and lifts syntax rewriting to a modularly-defined parsing problem. A notable effect is that we skirt the complexity and burden of defining additional translation layers between concrete user input templates and an underlying abstract syntax representation. We demonstrate that these ideas admit efficient and declarative rewrite templates across 12 languages, and validate effectiveness of our approach by producing correct and desirable lightweight transformations on popular real-world projects (over 50 syntactic changes produced by our approach have been merged into 40+). Our declarative rewrite patterns require an order of magnitude less code compared to analog implementations in existing, language-specific tools. 

   more » « less
5. Learning to reconstruct botanical trees from single images

   https://doi.org/10.1145/3478513.3480525  

   Li, Bosheng; Kałużny, Jacek; Klein, Jonathan; Michels, Dominik L.; Pałubicki, Wojtek; Benes, Bedrich; Pirk, Sören (December 2021, ACM Transactions on Graphics)

   We introduce a novel method for reconstructing the 3D geometry of botanical trees from single photographs. Faithfully reconstructing a tree from single-view sensor data is a challenging and open problem because many possible 3D trees exist that fit the tree's shape observed from a single view. We address this challenge by defining a reconstruction pipeline based on three neural networks. The networks simultaneously mask out trees in input photographs, identify a tree's species, and obtain its 3D radial bounding volume - our novel 3D representation for botanical trees. Radial bounding volumes (RBV) are used to orchestrate a procedural model primed on learned parameters to grow a tree that matches the main branching structure and the overall shape of the captured tree. While the RBV allows us to faithfully reconstruct the main branching structure, we use the procedural model's morphological constraints to generate realistic branching for the tree crown. This constraints the number of solutions of tree models for a given photograph of a tree. We show that our method reconstructs various tree species even when the trees are captured in front of complex backgrounds. Moreover, although our neural networks have been trained on synthetic data with data augmentation, we show that our pipeline performs well for real tree photographs. We evaluate the reconstructed geometries with several metrics, including leaf area index and maximum radial tree distances. 

   more » « less