Abstract Meaning Representation for Sembanking
What it is about
The paper presents Abstract Meaning Representation (AMR), a rooted, labelled graph for the meaning of whole sentences, meant to be easy for people to read and for programs to traverse. Concepts are English words, PropBank frames such as describe-01, or keywords for entities, dates and quantities. Edges are about a hundred relations, mostly PropBank arguments. The design aims at one graph per meaning, so that a passive, a nominal and a verbal form of the same sentence receive the same graph. Articles and tense are left out, and the authors state that AMR is English-biased and is not an interlingua. The paper describes a bank of several thousand sentences, annotated by trained annotators at about 7 to 10 minutes per sentence from scratch and 1 to 3 minutes for post-editing. It reports no parsing experiment.
Ideas
- I1. Function words are not concepts. AMR drops articles and tense from the graph and keeps only their semantic role. Keep determiners, tense and articles as features of the meta token (open problem 4), and not as tokens of their own.
- I2. Role-keyed merges: a concept merge must keep the argument roles of its members (arg0, arg1). A be/have merge fails this check, so a role test can block it without a list of banned pairs.
- I3. Score meta-worlds as sets of triples (concept, role, concept) with a Smatch-style F-score, reported beside exact match, so that a reading with one wrong relation is not scored as zero.
For Mova
The code is tools/translate/meta.rs and hebrew.rs. I1 and I2 address open problems 2 and 4. The graph format itself is not for Mova, but PropBank-keyed frames fit the frames that prag already builds, which makes the role check in I2 easy to write.
Library summary
Written for the Research library (claude-opus-5-5) from the paper text, in our own words: the license "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" does not allow redistributing the text itself. Original — aclanthology.org/W13-2322.pdf
Summary
The authors point to a problem: the semantic annotation of English texts is scattered across separate resources — named entities, coreference, temporal expressions, and so on. Each such resource has its own evaluation. Syntax, by contrast, has long had a single gold standard, such as the Penn Treebank. To fill this gap, the authors propose Abstract Meaning Representation (AMR). It is a language that describes the meaning of a whole sentence as a rooted, directed graph with labeled edges. For human readers, the graph is written in a PENMAN-based notation. AMR builds on PropBank frames. Instead of pronouns, it reuses variables. Sentences that differ only in form (for example, a verb versus a deverbal noun) receive the same representation. At the same time, AMR does not specify how exactly text is mapped to meaning, and it does not claim to be an interlingua, since it is oriented toward English. The outcome of the work is an annotation guideline for annotators, an editor, and a sembank of several thousand manually annotated sentences. Part of it can be downloaded free of charge; the rest is distributed by the LDC. The authors hope that such a resource will spur the development of statistical natural language understanding and generation, just as treebanks once spurred syntactic parsing.
Key points
- AMR has about 100 relations. Among them are PropBank frame arguments (:arg0–:arg5), general semantic relations (:location, :time, :cause…), and relations for quantities, dates, and lists (:op1–:op10). Each relation has an inverse (e.g., :arg0-of) and a reification (e.g., be-located-at-91).
- Negation is marked with :polarity, and modality with the concepts possible and obligate-01. Wh-questions use the concept amr-unknown, and other sentence types use the :mode relation. Deverbal nouns and adjectives are reduced to PropBank frames: investor → invest-01.
- Corpus composition. The authors jointly annotated 225 training sentences, 142 sentences from news, and 100 from the web. LDC annotators added 1546 sentences from "The Little Prince", 1328 and 1110 sentences from the web, 926 from Xinhua, and 214 from CCTV conversations. Some of these sets overlap with OntoNotes.
- In the AMR Editor, an annotator annotates a sentence from scratch in 7–10 minutes and corrects an existing annotation in 1–3 minutes. The annotation guideline runs to about 50 pages. Accuracy is evaluated with the Smatch metric (Cai & Knight, 2013).
- AMR is compared with GMB, UCCA, UNL, the Semantic Treebank, and the Prague Dependency Treebank. Future plans include machine translation via AMR (Chinese → AMR → English), disjunctive AMRs over HyTER paraphrase networks, as well as entity normalization and abstract frames.
Cite
@inproceedings{banarescu2013abstract,
title = {{Abstract Meaning Representation for Sembanking}},
author = {Banarescu, Laura and Bonial, Claire and Cai, Shu and Georgescu, Madalina and Griffitt, Kira and Hermjakob, Ulf and Knight, Kevin and Koehn, Philipp and Palmer, Martha and Schneider, Nathan},
booktitle = {Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse},
year = {2013},
url = {https://aclanthology.org/W13-2322/}
}References
- O. Abend and A. Rappoport. 2013. UCCA: A semantics-based grammatical annotation scheme. In Proc. IWCS. find
- C. Baker, C. Fillmore, and J. Lowe. 1998. The Berkeley FrameNet project. In Proc. COLING. find our card
- V. Basile, J. Bos, K. Evang, and N. Venhuizen. 2012a. Developing a large semantically annotated corpus. In Proc. LREC. Disjunctive AMR guideline: amr.isi.edu/damr.1.0.pdf V. Basile, J. Bos, K. Evang, and N. Venhuizen. 2012b. A platform for collaborative semantic annotation. In Proc. EACL demonstrations. find
- A. Böhmová, J. Hajič, E. Hajičová, and B. Hladká. 2003. The Prague dependency treebank. In Treebanks. Springer. find
- A. Butler and K. Yoshimoto. 2012. Banking meaning representations from treebanks. Linguistic Issues in Language Technology, 7. find
- S. Cai and K. Knight. 2013. Smatch: An accuracy metric for abstract meaning representations. In Proc. ACL. find
- D. Chiang, J. Andreas, D. Bauer, K. M. Hermann, B. Jones, and K. Knight. 2013. Parsing graphs with hyperedge replacement grammars. In Proc. ACL. find
- D. Davidson. 1969. The individuation of events. In N. Rescher, editor, Essays in Honor of Carl G. Hempel. D. Reidel, Dordrecht. find
- M. Dreyer and D. Marcu. 2012. Hyter: Meaning-equivalent semantics for translation evaluation. In Proc. NAACL. find
- B. Jones, J. Andreas, D. Bauer, K. M. Hermann, and K. Knight. 2012. Semantics-based machine translation with hyperedge replacement grammars. In Proc. COLING. find our card
- H. Kamp, J. Van Genabith, and U. Reyle. 2011. Discourse representation theory. In Handbook of philosophical logic, pages 125–394. Springer. find
- P. Kingsbury and M. Palmer. 2002. From TreeBank to PropBank. In Proc. LREC. find our card
- D. B. Lenat. 1995. Cyc: A large-scale investment in knowledge infrastructure. Communications of the ACM, 38(11). find
- R. Martins. 2012. Le Petit Prince in UNL. In Proc. LREC. find
- C. M. I. M. Matthiessen and J. A. Bateman. 1991. Text Generation and Systemic-Functional Linguistics. Pinter, London. find
- A. Meyers, R. Reeves, C. Macleod, R. Szekely, V. Zielinska, B. Young, and R. Grishman. 2004. The NomBank project: An interim report. In HLT-NAACL 2004 workshop: Frontiers in corpus annotation. find
- M. Palmer, D. Gildea, and P. Kingsbury. 2005. The Proposition Bank: An annotated corpus of semantic roles. Computational Linguistics, 31(1). find our card
- K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In ACL, Philadelphia, PA. find our card
- S. Pradhan, E. Hovy, M. Marcus, M. Palmer, L. Ramshaw, and R. Weischedel. 2007. Ontonotes: A unified relational semantic representation. International Journal of Semantic Computing (IJSC), 1(4). find
- D. Quernheim and K. Knight. 2012a. DAGGER: A toolkit for automata on directed acyclic graphs. In Proc. FSMNLP. find
- D. Quernheim and K. Knight. 2012b. Towards probabilistic acceptors and transducers for feature structures. In Proc. SSST Workshop. find
- K. Schuler. 2005. VerbNet: A broad-coverage, comprehensive verb lexicon. Ph.D. thesis, University of Pennsylvania. find
- S. Shieber, F. C. N. Pereira, L. Karttunen, and M. Kay. 1986. Compilation of papers on unification-based grammar formalisms. Technical Report CSLI-86-48, Center for the Study of Language and Information, Stanford, California. find
- H. Uchida, M. Zhu, and T. Della Senta. 1996. UNL: Universal Networking Language—an electronic language for communication, understanding and collaboration. Technical report, IAS/UNU Tokyo. find
- H. Uchida, M. Zhu, and T. Della Senta. 1999. A gift for a millennium. Technical report, IAS/UNU Tokyo. find
- N. Venhuizen, V. Basile, K. Evang, and J. Bos. 2013. Gamification for word sense labeling. In Proc. IWCS. find
- R. Weischedel, E. Hovy, M. Marcus, M. Palmer, R. Belvin, S. Pradhan, L. Ramshaw, and N. Xue. 2011. OntoNotes: A large training corpus for enhanced processing. In J. Olive, C. Christianson, and J. McCary, editors, Handbook of Natural Language Processing and Machine Translation. Springer. find
- Y. W. Wong and R. J. Mooney. 2006. Learning for semantic parsing with statistical machine translation. In Proc. HLT-NAACL. find