A More Precise Analysis of Punctuation for Broad-Coverage Surface Realization with CCG

What it is about

The paper gives punctuation a lexical analysis in a broad-coverage CCG realisation grammar. In CCGbank, punctuation marks carry categories taken from their tags and sit inside binary rules, which is enough for parsing but overgenerates in realisation: a comma category can wrap a sentence in commas or repeat itself. The authors discuss a syntactic-feature solution (Briscoe 1994) and reject it for CCG, because word order there is flexible. They choose semantic features instead. Every punctuation mark, except balancing marks, gets a category whose logical form carries a discourse function such as emphasis, elaboration or apposition. Appositives are split into balanced (sentence-medial) and unbalanced (sentence-final) cases by heuristic rules, and a rule-based filter bars illegal punctuation sequences and unbalanced apposition at realisation time. The corpus gained 89 punctuation categories in the development section. The authors report about a 9-point BLEU gain over the best oracle-n-gram baseline on non-blind data, a 40% rise in exact matches, and a blind-test BLEU gain above 5.5 points with over 96% coverage on section 23.

Ideas

• I1. A discourse-function feature on the meta token. Punctuation role (emphasis, elaboration, apposition) belongs on the meta token, and the encoder writes the mark from it. This is open problem 4 in its direct form. Derive the classes from the source side, count them per concept pair, and keep them as an enum.
• I2. Count balanced appositives in the target corpora. Measure how often a sentence-medial appositive is balanced in KJV and in WEB, and make the encoder's comma pairing a counted rule with a per-author delta.
• I3. A post-check filter with its own residual class. Reject illicit output sequences (doubled commas, a comma before a full clause) and count them as a separate class. Negative control: run the filter on shuffled tokens; it must flag most of them, or it is not a gate.
• I4. Split the punctuation residual into "mark chosen" and "mark placed". Which part dominates on KJV and WEB tells whether punctuation is a lexical or an ordering problem, and so whether it should be an author delta.

For Mova

The code is tools/translate/src/profile.rs for the encoder and the punctuation residual in tools/translate/src/measure.rs. The ideas touch open problem 4 (boundaries as features of meta tokens) and problem 8 (style deltas). The paper's filter is a rule, so it fits the deterministic design; its statistics do not transfer and must be counted on our own data.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work.

@inproceedings{white2008more,
  title = {{A More Precise Analysis of Punctuation for Broad-Coverage Surface Realization with CCG}},
  author = {White, Michael and Rajkumar, Rajakrishnan},
  booktitle = {Coling 2008: Proceedings of the workshop on Grammar Engineering Across Frameworks},
  year = {2008},
  url = {https://aclanthology.org/W08-1703/}
}

References

18 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Baldridge, Jason and Geert-Jan Kruijff. 2002. Coupling CCG and Hybrid Logic Dependency Semantics. In Proc. ACL-02. find
2. Baldridge, Jason. 2002. Lexically Specified Derivational Control in Combinatory Categorial Grammar. Ph.D. thesis, University of Edinburgh. find
3. Bayraktar, Murat, Bilge Say, and Varol Akman. 1998. An Analysis of English Punctuation: The Special Case of Comma. International Journal of Corpus Linguistics, 3(1):33–58. find
4. Boxwell, Stephen and Michael White. 2008. Projecting Propbank roles onto the CCGbank. In Proc. LREC-08. To appear. find
5. Briscoe, Ted. 1994. Parsing (with) punctuation. Technical report, Xerox, Grenoble, France. find
6. Doran, Christine. 1998. Incorporating Punctuation into the Sentence Grammar: A Lexicalized Tree Adjoining Grammar Perspective. Ph.D. thesis, University of Pennsylvania. find
7. Espinosa, Dominic, Michael White, and Dennis Mehay. 2008. Hypertagging: Supertagging for surface realization with CCG. In Proc. ACL-08:HLT. To appear. find
8. Forst, Martin and Ronald M. Kaplan. 2006. The importance of precise tokenizing for deep grammars. In Proc. LREC-06. find
9. Hockenmaier, Julia and Mark Steedman. 2007. CCGbank: A Corpus of CCG Derivations and Dependency Structures Extracted from the Penn Treebank. Computational Linguistics, 33(3):355–396. find our card
10. Hogan, Deirdre, Conor Cafferkey, Aoife Cahill, and Josef van Genabith. 2007. Exploiting multi-word units in history-based probabilistic generation. In Proc. EMNLP-CoNLL-07. find
11. Langkilde-Geary, Irene. 2002. An empirical verification of coverage and correctness for a general-purpose sentence generator. In Proc. INLG-02. find
12. Nakanishi, Hiroko, Yusuke Miyao, and Jun’ichi Tsujii. 2005. Probabilistic methods for disambiguation of an HPSG-based chart generator. In Proc. IWPT-05. find
13. Nunberg, Geoffrey. 1990. The Linguistics of Punctuation. CSLI Publications, Stanford, CA. find
14. Steedman, Mark and S. Prevost. 1994. Specifying intonation from context for speech synthesis. Speech Communication, 15(1–2):139–153. find
15. Steedman, Mark. 2000. The syntactic process. MIT Press, Cambridge, MA, USA. find
16. White, Michael, Rajakrishnan Rajkumar, and Scott Martin. 2007. Towards broad coverage surface realization with CCG. In Proc. of the Workshop on Using Corpora for NLG: Language Generation and Machine Translation (UCNLG+MT). find our card
17. White, Michael. 1995. Presenting punctuation. In Proceedings of the Fifth European Workshop on Natural Language Generation, pages 107–125. find
18. White, Michael. 2006. Efficient realization of coordinate structures in combinatory categorial grammar. Research on Language and Computation, 4(1):39– 75. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/white-2008-more-precise-analysis-punctuation-broad/