The Proposition Bank: An Annotated Corpus of Semantic Roles

By Martha Palmer, Paul R. Kingsbury, Daniel Gildea. 2005

2,691 citations recorded by the source corpus.

The Proposition Bank project takes a practical approach to semantic representation, adding a layer of predicate-argument information, or semantic role labels, to the syntactic structures of the Penn Treebank. The resulting resource can be thought of as shallow, in that it does not represent coreference, quantification, and many other higher-order phenomena, but also broad, in that it covers every instance of every verb in the corpus and allows representative statistics to be calculated. We discuss the criteria used to define the sets of semantic roles used in the annotation process and to analyze the frequency of syntactic/semantic alternations in the corpus. We describe an automatic system for semantic role tagging trained on the corpus and discuss the effect on its performance of various types of information, including a comparison of full syntactic parsing with a flat representation and the contribution of the empty trace categories of the treebank.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

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

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Deep Contextualized Word Representations (2018, methodology): “Labels consist of semantic roles from PropBank (Palmer et al., 2005) augmented with a BIO labeling scheme to represent argu-ment spans.”
Tailor: Generating and Perturbing Text with Semantic Controls (2021, methodology): “For other numbered arguments, we create human-readable labels by using argument functions included in the PropBank frame for the given predicate (Palmer et al., 2005).”
AllenNLP: A Deep Semantic Natural Language Processing Platform (2018, background): “AllenNLP includes reference implementations for several tasks, including: • Semantic Role Labeling (SRL) models recover the latent predicate argument structure of a sentence (Palmer et al., 2005).”
Modeling Biological Processes for Reading Comprehension (2014, methodology): “Like PropBank and BioProp, the argu-ment labels are not trigger speciﬁc.”
OntoNotes: The 90% Solution (2006, methodology): “The 1M word Penn Tree-bank II Wall Street Journal corpus has been successfully annotated with semantic argument structures for verbs and is now available via the Penn Linguistic Data Consortium as PropBank I (Palmer et al. , 2005).”
Cheap and Fast – But is it Good? Evaluating Non-Expert Annotations for Natural Language Tasks (2008, background): “Large scale annotation projects such as TreeBank (Marcus et al., 1993), PropBank (Palmer et al., 2005), TimeBank (Pustejovsky et al., 2003), FrameNet (Baker et al., 1998), SemCor (Miller et al., 1993), and others play an important role in natural language processing research, encouraging the…”
Question-Answer Driven Semantic Role Labeling: Using Natural Language to Annotate Natural Language (2015, background): “…is difﬁcult to formalize and fundamental aspects of the task vary across efforts, for example FrameNet (Baker et al., 1998) models a large set of interpretable thematic roles (AGENT, PATIENT, etc.) while PropBank (Palmer et al., 2005) uses a small set of verb-speciﬁc roles (ARG0, ARG1, etc.).”
Unsupervised Learning of Narrative Schemas and their Participants (2009, methodology): “Most previous work on unsupervised semantic role labeling assumes that the set of possible classes is very small (i.e, PropBank roles ARG0 and ARG1) and is known in advance.”
SemEval-2010 Task 8: Multi-Way Classification of Semantic Relations Between Pairs of Nominals (2009, background): “PropBank does use a small number of semantic roles, but these are again to be interpreted at the level of individual predicates, with little cross-predicate generalization.”
Understanding Factuality in Abstractive Summarization with FRANK: A Benchmark for Factuality Metrics (2021, methodology): “Our typology is theoretically grounded in frame semantics (Fillmore et al., 1976; Palmer et al., 2005) and linguistic discourse theory (Brown and Yule, 1983).”
Question Answering as Global Reasoning Over Semantic Abstractions (2018, methodology): “While it appears simple to use SRL systems for QA (Palmer, Gildea, and Kingsbury 2005), this has found limited success (Kaisser and Webber 2007; Pizzato and Moll ´ a 2008; Moreda et al. 2011).”
Annotating Twitter Data from Vulnerable Populations: Evaluating Disagreement Between Domain Experts and Graduate Student Annotators (2018, background): “Machine learning and natural language research has long relied on the robust annotation of social media data to examine and predict myriad human phenomenon [10, 12, 14].”
Introduction to the CoNLL-2005 Shared Task: Semantic Role Labeling (2005, methodology): “The Proposition Bank (PropBank) (Palmer et al., 2005) annotates the Penn TreeBank with verb argu-ment structure.”
Deep Graph Convolutional Encoders for Structured Data to Text Generation (2018, methodology): “Here we focus on the deep task where the input is a semantic dependency graph that represents a target sentence using predicate-argument structures (NomBank; (Meyers et al., 2004), PropBank; (Palmer et al., 2005)).”
The CoNLL 2008 Shared Task on Joint Parsing of Syntactic and Semantic Dependencies (2008, methodology): “The PropBank annotation (Palmer et al., 2005) classiﬁes the arguments of all the main verbs in the Penn Treebank corpus, other than be .”
SemEval-2007 Task-17: English Lexical Sample, SRL and All Words (2007, methodology): “Traditionally, SRL systems have been trained on either the PropBank corpus (Palmer et al., 2005) – for two years, the CoNLL workshop (Carreras and Màrquez, 2004; Carreras and Màrquez, 2005) has made this their shared task, or the FrameNet corpus – Senseval-3 used this for their shared task…”
Efficient Inference in Large Conditional Random Fields (2006, methodology): “This model was trained on the proposition-bank corpus [10], using the CoNLL 2005 shared task dataset [11].”