The Penn Treebank: Annotating Predicate Argument Structure

By Mitchell P. Marcus, Grace Kim, Mary Ann Marcinkiewicz, R. MacIntyre, Ann Bies, Mark Ferguson, Karen Katz, Britta Schasberger. 1994

The Penn Treebank has recently implemented a new syntactic annotation scheme, designed to highlight aspects of predicate-argument structure. This paper discusses the implementation of crucial aspects of this new annotation scheme. It incorporates a more consistent treatment of a wide range of grammatical phenomena, provides a set of coindexed null elements in what can be thought of as "underlying" position for phenomena such as wh-movement, passive, and the subjects of infinitival constructions, provides some non-context free annotational mechanism to allow the structure of discontinuous constituents to be easily recovered, and allows for a clear, concise tagging system for some semantic roles.

Subfields: Information Extraction

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Information Extraction

Citations in the corpus

GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers (2022, methodology): “Table 5 summarizes results across Wikitext-2, PTB, C4.”
Efficient Neural Architecture Search via Parameter Sharing (2018, background): “On Penn Treebank, our method achieves a test per-plexity of 55 .”
AutoML: A Survey of the State-of-the-Art (2019, methodology): “Although these two works have achieved state-of-the-art results of datasets including CIFAR10 and Penn Treebank (PTB [100]), [5] took 28 days and 800 K40 GPUs to search for the best-performing architecture, which is unaffordable for individual researchers and even companies, MetaQNN [16] took 10 day”
Neural Architecture Optimization (2018, background): “Furthermore, on PTB dataset we achieve 56 .”
Evaluating the Search Phase of Neural Architecture Search (2019, methodology): “We report results on PTB us-ing mean validation perplexity (the lower, the better) and on CIFAR-10 using mean top 1 accuracy.”
Overcoming Multi-Model Forgetting (2019, positive): “Our final results on the best architecture found by the search confirm that limiting multi-model forgetting yields better results and better convergence for both language modeling (on the PTB dataset (Marcus et al., 1994)) and image classification (on the CIFAR10 dataset (Krizhevsky et al.”
Multiple Interactions Made Easy (MIME): Large Scale Demonstrations Data for Imitation (2018, background): “Inspired from the successes in large scale computer vision datasets [1, 12], natural language datasets [13] and reinforcement learning frameworks [14, 15], the last few years has seen a growing interest in large scale robotics [2, 3].”
NSGA-NET: A Multi-Objective Genetic Algorithm for Neural Architecture Search (2018, methodology): “PTB refers to the Penn Treebank Marcus et al. (1994) dataset.”
Head-Driven Statistical Models for Natural Language Parsing (2003, background): “See Marcus et al. (1994) for an explanation of what these tags signify.”
What’s in an Embedding? Analyzing Word Embeddings through Multilingual Evaluation (2015, methodology): “The Penn Treebank (Marcus et al., 1994), converted using the LTH converter (Johansson and Nugues, 2007), was used as the corresponding manually annotated resource.”
Unsupervised Question Answering by Cloze Translation (2019, methodology): “It is worth noting that to attain our best performance, we require the use of both an NER system, indirectly using labelled data from OntoNotes 5, and a constituency parser for extracting sub-clauses, trained on the Penn Treebank (Marcus et al., 1994).”
Efficient Inference in Large Conditional Random Fields (2006, methodology): “We used the Penn Treebank III [6], training on sections 2–21, using section 24 for development and section 23 for testing.”
Domain-Independent Dominance of Adaptive Methods (2019, methodology): “We now consider a task where state-of-the-art results are achieved by adaptive methods with small values for ǫ and where SGD has little success: character-level language modelling with LSTMs [17] on the Penn Treebank dataset [29, 31].”
Square One Bias in NLP: Towards a Multi-Dimensional Exploration of the Research Manifold (2022, background): “For many years, the English Penn Treebank (Marcus et al., 1994) was an integral part of the SQUARE ONE of NLP.”