[{"_1":2,"_161":-5,"_162":-5},"loaderData",{"_3":4,"_17":18,"_25":26,"_140":141},"root",{"_5":-5,"_6":-7,"_7":8,"_9":-7,"_10":11,"_12":-7,"_13":14,"_15":16},"currentUser","primaryEmail","siteVerification","6LeIxAcTAAAAAJcZVRqyHh71UMIEGNQ_MXjiZKhI","toast","enableDev",false,"incidentStatus","theme","system","contrast","normal","routes/_layout",{"_19":-5,"_20":21,"_22":21,"_23":11,"_24":21},"username","isAskGordonEnabled",true,"isUsingNewTemplate","showMinimalLanding","showHoneypotTrap","routes/_layout.r.$namespace.$reponame",{"_27":28,"_29":-5,"_30":11,"_31":11,"_32":33,"_34":35,"_36":-5,"_37":11,"_38":39,"_88":-5,"_89":90,"_129":-5,"_130":11,"_131":21,"_132":11,"_133":-5,"_134":11,"_135":-5,"_136":137,"_138":91,"_139":-7},"canonicalUrl","https://hub.docker.com/r/masashiy/ccg2lambda/tags","mcpServer","canSeeDockerfile","canSeeAutobuilds","namespace","masashiy","reponame","ccg2lambda","cloudBuildSource","isAutobuildsAvailable","repo",{"_40":41,"_42":43,"_45":46,"_47":48,"_49":11,"_50":51,"_52":53,"_54":55,"_57":35,"_32":33,"_58":59,"_60":44,"_61":62,"_63":64,"_65":66,"_67":-5,"_68":62,"_69":70,"_71":11,"_72":33,"_73":74,"_79":21,"_80":81,"_85":33,"_86":87},"categories",[],"content_types",[44],"image","date_registered","2018-11-06T23:22:24.866596Z","description","ccg2lambda with several CCG parsers","is_private","last_updated","2019-09-22T05:32:01.195953Z","last_modified","2024-10-16T13:48:34.145251Z","media_types",[56],"application/vnd.docker.container.image.v1+json","name","pull_count",932,"repository_type","star_count",0,"status",1,"status_description","active","source","collaborator_count","full_description","# ccg2lambda: composing semantic representations guided by CCG derivations\n\nThis is a tool to derive formal semantic representations of\nnatural language sentences given CCG derivation trees and semantic templates.\n\n## Installation\n\nIn order to run most part of this software, it is necessary to install python3,\nnltk 3, lxml, simplejson and yaml python libraries. I recommend to use a python virtual environment,\nand install the packages in the virtual environment with `pip`:\n\n```bash\nsudo apt-get install python3-dev\nsudo apt-get install python-virtualenv\nsudo apt-get install libxml2-dev libxslt1-dev\ngit clone https://github.com/mynlp/ccg2lambda.git\ncd ccg2lambda\nvirtualenv --no-site-packages --distribute -p /usr/bin/python3 py3\nsource py3/bin/activate\npip install lxml simplejson pyyaml -I nltk==3.0.5\n```\n\nYou also need to install WordNet:\n\n```bash\npython -c \"import nltk; nltk.download('wordnet')\"\n```\n\nTo ensure that all software is working as expected, you can run the tests:\n\n```bash\npython scripts/run_tests.py\n```\n(all tests should pass, except a few expected failures).\n\nYou also need to install the [Coq Proof Assistant](https://coq.inria.fr/) that we use for automated reasoning. In Ubuntu, you can install it by:\n\n```bash\nsudo apt-get install coq\n```\n\nThen, compile the coq library that contains the axioms:\n\n```bash\ncoqc coqlib.v\n```\n\nOur system assigns semantics to CCG structures. At the moment, we support C&C for English, and Jigg for Japanese.\n\n### Installing [C&C parser](http://www.cl.cam.ac.uk/~sc609/candc-1.00.html) (for English)\n\nYou can download and install the C&C syntactic parser by running the following script\nfrom the ccg2lambda directory:\n\n```bash\n./en/install_candc.sh\n```\n\nIf that fails, you may succeed by following [these alternative instructions](https://github.com/valeriobasile/learningbyreading#installation-of-the-cc-tools-and-boxer), in which case you need to manually create a file `en/candc_location.txt` with the path to the C&C parser:\n\n```bash\necho \"/path/to/candc-1.00/\" > en/candc_location.txt\n```\n\n### Installing [Jigg parser](https://github.com/mynlp/jigg) (for Japanese)\n\nSimply do:\n\n```bash\n./ja/download_dependencies.sh\n```\n\nThe command above will download Jigg, its models, and create the file `ja/jigg_location.txt` where the path to Jigg is specified. That is all.\n\n## Using the Semantic Parser\n\nLet's assume that we have a file `sentences.txt` with one sentence per line,\nand that we want to semantic parse those sentences. Here is the content of\nmy file:\n\n```\nAll women ordered coffee or tea.\nSome woman did not order coffee.\nSome woman ordered tea.\n```\n\nAnd we want to obtain a symbolic semantic representation such as:\n\n```\nforall x. (woman(x) -> exists y. ((tea(y) \\/ coffee(y)) /\\ order(x, y)))\nexists x. (woman(x) /\\ -exists y. (cofee(y) /\\ order(x, y)))\nexists x. (woman(x) /\\ exists y. (tea(y) /\\ order(x, y)))\n```\n\nFirst we need to obtain the CCG derivations (parse trees) of the sentences\nin the text file using C&C and convert its XML format into Jigg's XML format:\n\n```bash\ncat sentences.txt | sed -f en/tokenizer.sed > sentences.tok\n/path/to/candc-1.00/bin/candc --models /path/to/candc-1.00/models --candc-printer xml --input sentences.tok > sentences.candc.xml\npython en/candc2transccg.py sentences.candc.xml > sentences.xml\n```\n\nThen, we are ready to obtain the semantic representations by using semantic\ntemplates and the CCG derivations obtained above:\n\n```bash\npython scripts/semparse.py sentences.xml en/semantic_templates_en_emnlp2015.yaml sentences.sem.xml\n```\n\nThe semantic representations are in the `sentences.sem.xml` file,\nwhere a new XML node `` has been added with as many child nodes\nas the CCG structure. Each semantic span has the logical representation\nobtained up to that span. The root span has the logical representation\nof the whole sentence. Here there is an excerpt of the semantics XML node\nof the last sentence:\n\n```xml\n\n \n \n \n \n \n \n \n \n \n \n\n```\n\nThe `sem` attribute contains the logical formulas, and the `type` attributes\nthe types of the predicates (types only appear at the leaves).\n\n## Using a prover (Coq) for recognizing textual entailment\n\nWe believe that the semantic representations above can be used\nfor several NLP tasks. We have been using them so far\nfor recognizing textual entailment. For this purpose,\nwe assume that all sentences in the file are premises,\nexcept the last one, which is the conclusion.\n\nTo build a theorem out of those logical representations,\npipe it to a theorem prover (Coq) and judge the entailment\nrelation, you can run the following command:\n\n```bash\npython scripts/prove.py sentences.sem.xml --graph_out graphdebug.html\n```\n\nThat command will output `yes` (entailment relation - the conclusion\ncan be proved given the premises), `no` (contradiction - the negated\nconclusion can be proved), `unknown` (otherwise).\n\nIf the parsing process and theorem proving succeeded,\ngraphdebug.html will have a graphical representation\nof the CCG trees, augmented with logical formulas at\nevery node below the syntactic category. The script\nthat pipes the theorem to Coq is also displayed at\nthe bottom. If the semantic parsing or prover fails,\ngraphdebug.html may contain plain debugging information\n(e.g. python error messages, etc.). Here is the `graphdebug.html`\nof the example above:\n\n![alt text](./doc/images/graphdebug.png \"Visualization of semantic parser and prover\")\n\n## Visualization\n\nIt is also possible to visualize CCG trees, either before\nor after augmenting them with semantic representations.\nFor example, to visualize the CCG trees only (without\nsemantic representations):\n\n```bash\npython scripts/visualize.py sentences.xml > sentences.html\n```\n\nand then open the file `sentences.html` with your favourite web browser.\nYou should be able to see something like this:\n\n![alt text](./doc/images/ccg_html.png \"Visualization of CCG trees (without semantic representations)\")\n\n## Reproducibility\n\nIf you wish to reproduce our reported results, please follow the instructions below:\n\n* [Experiments on FraCaS at EMNLP 2015](en/fracas.md)\n* [Experiments on JSeM at EMNLP 2016](ja/jsem.md)\n* [Experiments on SICK at EACL 2017](en/sick.md)\n* [Experiments on STS for SICK and MSRvid at EMNLP 2017](en/emnlp2017_sts.md)\n* [Experiments on SICK at NAACL-HLT 2018](https://github.com/verypluming/ccg2lambda/blob/naacl/en/naacl2018.md)\n\n## Interpreting (and writing your own) semantic templates.\n\nYou can find one of our semantic templates in `en/semantic_templates_en.yaml`. Here are some notes:\n\n1. Each Yaml block is a rule. If the rule matches the attributes of a CCG node, then the semantic template specified by \"semantics\" is applied. It is possible to write an arbitrary set of field names and their values. For example, you could specify the \"category\" of the CCG node and the surface \"surf\" form of a word (in case the CCG node is a leaf). Only the \"category\" field and the \"semantics\" field are compulsory.\n2. If you underspecify the characteristics of a CCG node, the semantic rule will match more general CCG nodes. It is also possible to underspecify the features of syntactic categories.\n3. If more than one rule applies, only the last one in the file will be applied.\n4. Any CCG node can be matched by the semantic rules.\n 1. If the rule matches a CCG leaf node, then the base form (or surface form, if base not present) will be passed as the first argument to the lambda expression of the semantic template (specified by \"semantics\" field in Yaml rule).\n 2. If the rule matches a CCG node with only one child, then the semantic expression of the child will be passed as the first argument to the semantic template assigned to this CCG node.\n 3. If the rule matches a CCG node with two children, then the semantic expressions of the left and right children will be passed as the first and second argument to the semantic template of this CCG node, respectively.\n5. Rules that are intended to match inner nodes of the CCG derivation need to specify either:\n 1. A \"rule\" attribute, specifying what type of combination operation is there.\n 2. A \"child\" attribute, specifying at least one attribute of one child (see next point).\n6. You can specify in a semantic rule the characteristics of the children at the current CCG node. To that purpose, simply prefix the attribute name with \"child0\\_\" or \"child1\\_\" of the left or right child, respectively. E.g. if you want to specify the syntactic category of the left child, then you call such attribute in the semantic rule as \"child0\\_category\". You can still underspecify the attributes of the children. If the CCG node only has one child, then it is considered a left child (and its attributes are specified using \"child0\\_\" as a prefix).\n\n# Citing this work\n\nIf you use this software or the semantic templates for your work, please consider citing it.\n\n## The system - software for compositional semantics:\n\n* Pascual Martínez-Gómez, Koji Mineshima, Yusuke Miyao, Daisuke Bekki. ccg2lambda: A Compositional Semantics System. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics - System Demonstrations, pages 85–90, Berlin, Germany, August 7-12, 2016. [pdf](https://aclweb.org/anthology/P/P16/P16-4015.pdf)\n\n```\n@InProceedings{pascual-EtAl:2016:ACL-2016-System-Demonstrations,\n author = {Mart\\'{i}nez-G\\'{o}mez, Pascual and Mineshima, Koji and Miyao, Yusuke and Bekki, Daisuke},\n title = {ccg2lambda: A Compositional Semantics System},\n booktitle = {Proceedings of ACL 2016 System Demonstrations},\n month = {August},\n year = {2016},\n address = {Berlin, Germany},\n publisher = {Association for Computational Linguistics},\n pages = {85--90},\n url = {https://aclweb.org/anthology/P/P16/P16-4015.pdf}\n}\n```\n\n## A system to inject phrasal axioms on-demand (phrase abduction):\n\n * Hitomi Yanaka, Koji Mineshima, Pascual Martínez-Gómez and Daisuke Bekki. Acquisition of Phrase Correspondences using Natural Deduction Proofs. Proceedings of 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, New Orleans, Louisiana, 1-6 June 2018. [arXiv](https://arxiv.org/pdf/1804.07656.pdf)\n\n ```\n @InProceedings{yanaka-EtAl:2018:NAACL-HLT,\n author = {Yanaka, Hitomi and Mineshima, Koji and Mart\\'{i}nez-G\\'{o}mez, Pascual and Bekki, Daisuke},\n title = {Acquisition of Phrase Correspondences using Natural Deduction Proofs},\n booktitle = {Proceedings of 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},\n month = {June},\n year = {2018},\n address = {New Orleans, Louisiana},\n publisher = {Association for Computational Linguistics},\n }\n ```\n\n## A system to compute Semantic Sentence Similarity:\n\n* Hitomi Yanaka, Koji Mineshima, Pascual Martínez-Gómez and Daisuke Bekki. Determining Semantic Textual Similarity using Natural Deduction Proofs. Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, Copenhagen, Denmark, 7-11 September 2017. [arXiv](https://arxiv.org/pdf/1707.08713.pdf)\n\n```\n@InProceedings{yanaka-EtAl:2017:EMNLP,\n author = {Yanaka, Hitomi and Mineshima, Koji and Mart\\'{i}nez-G\\'{o}mez, Pascual and Bekki, Daisuke},\n title = {Determining Semantic Textual Similarity using Natural Deduction Proofs},\n booktitle = {Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing},\n month = {September},\n year = {2017},\n address = {Copenhagen, Denmark},\n publisher = {Association for Computational Linguistics},\n}\n```\n\n\n## A mechanism to inject axioms on-demand:\n\n* Pascual Martínez-Gómez, Koji Mineshima, Yusuke Miyao, Daisuke Bekki. On-demand Injection of Lexical Knowledge for Recognising Textual Entailment. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics, pages 710-720, Valencia, Spain, 3-7 April, 2017. [pdf](http://www.aclweb.org/anthology/E17-1067)\n\n```\n@InProceedings{martinezgomez-EtAl:2017:EACLlong,\n author = {Mart\\'{i}nez-G\\'{o}mez, Pascual and Mineshima, Koji and Miyao, Yusuke and Bekki, Daisuke},\n title = {On-demand Injection of Lexical Knowledge for Recognising Textual Entailment},\n booktitle = {Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers},\n month = {April},\n year = {2017},\n address = {Valencia, Spain},\n publisher = {Association for Computational Linguistics},\n pages = {710--720},\n url = {http://www.aclweb.org/anthology/E17-1067}\n}\n```\n\n## The English semantic model declared as semantic templates:\n\n* Koji Mineshima, Pascual Martínez-Gómez, Yusuke Miyao, Daisuke Bekki. Higher-order logical inference with compositional semantics. Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2055–2061, Lisbon, Portugal, 17-21 September 2015. [pdf](http://www.aclweb.org/anthology/D15-1244)\n\n```\n@InProceedings{mineshima-EtAl:2015:EMNLP,\n author = {Mineshima, Koji and Mart\\'{i}nez-G\\'{o}mez, Pascual and Miyao, Yusuke and Bekki, Daisuke},\n title = {Higher-order logical inference with compositional semantics},\n booktitle = {Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing},\n month = {September},\n year = {2015},\n address = {Lisbon, Portugal},\n publisher = {Association for Computational Linguistics},\n pages = {2055--2061},\n url = {http://aclweb.org/anthology/D15-1244}\n}\n```\n\n## The Japanese semantic model declared as semantic templates:\n\n* Koji Mineshima, Ribeka Tanaka, Pascual Martínez-Gómez, Yusuke Miyao, Daisuke Bekki. Building compositional semantics and higher-order inference system for wide-coverage Japanese CCG parsers. Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2236-2242, Austin, Texas, 1-5 November 2016. [pdf](http://aclweb.org/anthology/D16-1242)\n\n```\n@InProceedings{D16-1242,\n author = \t\"Mineshima, Koji\n\t\tand Tanaka, Ribeka\n\t\tand Mart{\\'i}nez-G{\\'o}mez, Pascual\n\t\tand Miyao, Yusuke\n\t\tand Bekki, Daisuke\",\n title = \t\"Building compositional semantics and higher-order inference system for a wide-coverage {J}apanese {CCG} parser\",\n booktitle = \t\"Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing\",\n year = \t\"2016\",\n publisher = \t\"Association for Computational Linguistics\",\n pages = \t\"2236--2242\",\n location = \t\"Austin, Texas\",\n url = \t\"http://aclweb.org/anthology/D16-1242\"\n}\n```\n","has_starred","hub_user","immutable_tags_settings",{"_75":11,"_76":77},"enabled","rules",[78],".*","is_automated","permissions",{"_82":21,"_83":11,"_84":11},"read","write","admin","user","storage_size",4589753942,"orgData","recentTags",[91],{"_92":44,"_93":94,"_95":96,"_97":98,"_99":100,"_50":118,"_119":94,"_120":33,"_121":56,"_57":122,"_123":124,"_125":109,"_126":118,"_127":66,"_128":21},"content_type","creator",4522906,"full_size",2246207899,"id",39378883,"images",[101],{"_102":103,"_104":105,"_106":107,"_108":109,"_110":111,"_112":113,"_114":107,"_115":-5,"_116":96,"_63":66,"_117":-5},"architecture","amd64","digest","sha256:c8857c7a60aa08c2964613609d946ec6c94a813746e5e62232f0260d0b9c0a04","features","","last_pulled","2026-10-02T02:04:31.025463235Z","last_pushed","2019-09-22T05:32:01Z","os","linux","os_features","os_version","size","variant","2019-09-22T05:32:00.888153Z","last_updater","last_updater_username","media_type","latest","repository",6253887,"tag_last_pulled","tag_last_pushed","tag_status","v2","sbxKitMetadata","isPotentialSpam","isHubSbxKitsEnabled","showSbxLogoOverride","sbxTrademarkText","displaySecurityMetadata","tagAttestations","tagsHaveAttestations",[],"selectedTag","attestationsUrl","routes/_layout.r.$namespace.$reponame.tags",{"_38":142,"_149":150,"_136":159,"_160":11},{"_40":143,"_42":144,"_45":46,"_47":48,"_49":11,"_50":51,"_52":53,"_54":145,"_57":35,"_32":33,"_58":59,"_60":44,"_61":62,"_63":64,"_65":66,"_67":-5,"_68":62,"_69":70,"_71":11,"_72":33,"_73":146,"_79":21,"_80":148,"_85":33,"_86":87},[],[44],[56],{"_75":11,"_76":147},[78],{"_82":21,"_83":11,"_84":11},"tags",{"_151":64,"_152":-5,"_153":-5,"_154":155},"count","next","previous","results",[156],{"_92":44,"_93":94,"_95":96,"_97":98,"_99":157,"_50":118,"_119":94,"_120":33,"_121":56,"_57":122,"_123":124,"_125":109,"_126":118,"_127":66,"_128":21},[158],{"_102":103,"_104":105,"_106":107,"_108":109,"_110":111,"_112":113,"_114":107,"_115":-5,"_116":96,"_63":66,"_117":-5},[],"canDeleteTags","actionData","errors"]

masashiy/ccg2lambda

Tags for masashiy/ccg2lambda