{"id":841,"date":"2015-10-23T09:53:26","date_gmt":"2015-10-23T14:53:26","guid":{"rendered":"http:\/\/haleyai.com\/wordpress\/?p=841"},"modified":"2018-08-05T05:03:17","modified_gmt":"2018-08-05T10:03:17","slug":"robust-inference-and-slacker-semantics","status":"publish","type":"post","link":"http:\/\/haleyai.com\/wordpress\/2015\/10\/23\/robust-inference-and-slacker-semantics\/","title":{"rendered":"Robust Inference and Slacker Semantics"},"content":{"rendered":"<p>In preparing for some natural language generation<a href=\"#_ftn1\"><sup><\/sup><sup>[1]<\/sup><\/a>, I came across some work on natural logic<a href=\"#_ftn2\"><sup><\/sup><sup>[2]<\/sup><\/a><a href=\"#_ftn3\"><sup><\/sup><sup>[3]<\/sup><\/a> and reasoning by textual entailment<a href=\"#_ftn4\"><sup><\/sup><sup>[4]<\/sup><\/a> (<a href=\"http:\/\/www.nist.gov\/tac\/2011\/RTE\/index.html\">RTE<\/a>) by <a href=\"http:\/\/richard.bergmair.eu\/\">Richard Bergmair<\/a> in his PhD at Cambridge:<\/p>\n<ul>\n<li><a href=\"http:\/\/richard.bergmair.eu\/pub\/thesis.pdf\">Monte Carlo Semantics: Robust Inference and Logical Pattern Processing with Natural Language Text<\/a><\/li>\n<\/ul>\n<p>The work he describes overlaps our approach to robust inference from the deep, variable-precision semantics that result from linguistic analysis and disambiguation using the English Resource Grammar (<a href=\"http:\/\/www.delph-in.net\/erg\/\">ERG<\/a>) and the <a href=\"http:\/\/haleyai.com\/wordpress\/2013\/05\/29\/translating-english-into-logic-using-the-linguist\/\">Linguist<\/a>&#x2122;.<\/p>\n<p><!--more--><\/p>\n<p>Mr. Bergmair\u2019s semantic logic project has 2 components:<\/p>\n<ol>\n<li>A Python platform for experimentation with semantics:<br \/>\ni.e., software for converting the minimal recursion semantics (<a href=\"https:\/\/en.wikipedia.org\/wiki\/Minimal_recursion_semantics\">MRS<\/a>) produced using the ERG into first-order logic<\/li>\n<li>A textual entailment engine using Monte Carlo techniques on the first-order predicate calculus (FOPS) produced above<\/li>\n<\/ol>\n<p>The following comments are particularly interesting:<\/p>\n<ul>\n<li>With the scoping machinery and the first-order approximation in place, PyPES&#x2122; makes it possible to translate text into formulae of FOPC. This is what <a href=\"http:\/\/svn.ask.it.usyd.edu.au\/trac\/candc\/wiki\/boxer\">Boxer<\/a> does for <a href=\"http:\/\/en.wikipedia.org\/wiki\/Combinatory_categorial_grammar\">CCG<\/a> and what <a href=\"http:\/\/en.wikipedia.org\/wiki\/Glue_Semantics\">Glue Semantics<\/a> does for <a href=\"http:\/\/en.wikipedia.org\/wiki\/Lexical_functional_grammar\">LFG<\/a>.\n<ul>\n<li>\u2026the main problem with Boxer and glue semantics is their strong commitment throughout to classical bivalent logic, which is limited in its ability to represent natural language semantics.<\/li>\n<li>FOPC lacks some kinds of expressive power that are important for natural language, such as quantifiers like <em><\/em><em>most<\/em> as well as weakening and strengthening modifiers like <em><\/em><em>very<\/em>.<\/li>\n<li>The straightforward logical encoding used by Boxer and glue semantics leads to over-commitment in some places, for example forcing strictly recursive quantifier scopings when little or nothing is known about the scopings from the natural language input<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>Parenthetically, CCG is <a href=\"https:\/\/en.wikipedia.org\/wiki\/Combinatory_categorial_grammar\">combinatorial categorical grammar<\/a> (spelled in various ways) and LFG is <a href=\"https:\/\/en.wikipedia.org\/wiki\/Lexical_functional_grammar\">lexical function grammar<\/a>.\u00a0 LFG and glue semantics were developed at Xerox PARC before being commercialized at <a href=\"https:\/\/en.wikipedia.org\/wiki\/Powerset_%28company%29\">Powerset<\/a> and subsequently acquired by Microsoft. \u00a0\u00a0The ERG is a <a href=\"https:\/\/en.wikipedia.org\/wiki\/Head-driven_phrase_structure_grammar\">head-driven phrase structure grammar<\/a> (HPSG).\u00a0 All of the parsing systems that produce deeper, albeit under-specified semantic representation<a href=\"#_ftn5\"><sup><\/sup><sup>[5]<\/sup><\/a>, including HPSG, CCG, and LFG, are <a href=\"https:\/\/www.google.com\/url?sa=t&amp;rct=j&amp;q=&amp;esrc=s&amp;source=web&amp;cd=1&amp;ved=0CB4QFjAAahUKEwiB5-L-4NjIAhUJOD4KHeHDA6w&amp;url=http%3A%2F%2Fwww.cs.haifa.ac.il%2F~shuly%2Fmalta-slides.pdf&amp;usg=AFQjCNEuyGfJV_qouXQPbGak7tPzcGlfyg&amp;cad=rja\">unification grammars<\/a>.<\/p>\n<p>I have long followed <a href=\"http:\/\/www.rug.nl\/staff\/johan.bos\/research\">Johan Bos<\/a> work on under-specified representation and his use of <a href=\"https:\/\/en.wikipedia.org\/wiki\/Lambda_calculus\">lambda calculus<\/a> and Prolog (Boxer) and CCG.\u00a0 For example, visit the <a href=\"http:\/\/gmb.let.rug.nl\/about.php\">Groningen Meaning Bank<\/a>.\u00a0 Nonetheless, Mr. Bergmair\u2019s criticism is fair and reasonable.\u00a0 In our work at Vulcan<a href=\"#_ftn6\"><sup><\/sup><sup>[6]<\/sup><\/a>, the <a href=\"https:\/\/en.wikipedia.org\/wiki\/Well-founded_semantics\">well-founded semantics<\/a> proved critical, for example. \u00a0<a href=\"http:\/\/plato.stanford.edu\/entries\/generalized-quantifiers\/\">Generalized quantification<\/a> was also critical, including the use of a reasoning engine that could support such quantification defeasibly<a href=\"#_ftn7\"><sup><\/sup><sup>[7]<\/sup><\/a>.<\/p>\n<p>Mr.Bergmair continues his comment regarding strict resolution of all scope ambiguities by citing the use of <a href=\"http:\/\/www.aclweb.org\/anthology\/E\/E09\/E09-1001.pdf\">slacker semantics<\/a> rather than resolving ambiguities of scope.\u00a0 Implicit in this statement is that the first component is not typically used to resolve all scopal ambiguities.<\/p>\n<p>I should note that Mr. Bergmair does not discuss the resolution of grammatical ambiguity.\u00a0 For short sentences, as in most of his examples, grammatical ambiguity is low.\u00a0 But it is combinatoric and becomes significant for sentences over 10 words long.\u00a0 Presumably, Mr. Bergmair works from the best ranking parse. \u00a0However, the best ranking parse rarely includes the intended semantics for sentences significantly more than 10 words in length.<a href=\"#_ftn8\"><sup><\/sup><sup>[8]<\/sup><\/a><\/p>\n<p>I am not surprised that Mr. Bergmair does not address grammatical ambiguity and deemphasizes resolution of scopal ambiguities since to do so can be difficult without a user interface such as in the Linguist.\u00a0 Given an RTE focus this is a reasonable position since there is no human assistance as in a cognitive-computing approach.\u00a0 However, for sentences of moderate length or longer, ambiguity of grammar, let alone of logic will produce combinatorial ambiguities of entailment that even a Monte Carlo approach may not be able to address.\u00a0 If this is the case, then slacker semantics (i.e., heuristic reasoning from under-specified semantics) will have to be reduced significantly, perhaps close to the point of elimination of ambiguity.<\/p>\n<p>This is precisely what we emphasize with the Linguist.\u00a0 Consequently, Mr. Bergmair\u2019s approach and others we are pursuing will result in more robust, deep, and precise reasoning.<\/p>\n<hr size=\"1\" \/><a href=\"#_ftnref1\"><sup><\/sup><sup>[1]<\/sup><\/a><strong><a href=\"https:\/\/www.google.com\/url?sa=t&amp;rct=j&amp;q=&amp;esrc=s&amp;source=web&amp;cd=8&amp;ved=0CFsQFjAHahUKEwjCgZ_J3NjIAhXKHT4KHQHfA9w&amp;url=http%3A%2F%2Fcs.jhu.edu%2F%7Exuchen%2Fpaper%2FYao2010Master.pdf&amp;usg=AFQjCNHWBIiLiXb69ExVQd_dYnnBZOd4gw&amp;cad=rja\">Question Generation with Minimal Recursion Semantics<\/a><\/strong><\/p>\n<p><a href=\"#_ftnref2\"><sup><\/sup><sup>[2]<\/sup><\/a> http:\/\/maartens.home.xs4all.nl\/philosophy\/Dictionary\/N\/Natural%20Logic.htm<\/p>\n<p><a href=\"#_ftnref3\"><sup><\/sup><sup>[3]<\/sup><\/a> http:\/\/nlp.stanford.edu\/projects\/natlog.shtml<\/p>\n<p><a href=\"#_ftnref4\"><sup><\/sup><sup>[4]<\/sup><\/a> http:\/\/www.aclweb.org\/aclwiki\/index.php?title=Textual_Entailment_References<\/p>\n<p><a href=\"#_ftnref5\"><sup><\/sup><sup>[5]<\/sup><\/a> <strong><a href=\"http:\/\/link.springer.com\/content\/pdf\/10.1007\/978-1-4020-5958-2_4.pdf\">Semantic underspecification: Which technique for what purpose?<\/a><\/strong><\/p>\n<p><a href=\"#_ftnref6\"><sup><\/sup><sup>[6]<\/sup><\/a> <a href=\"http:\/\/haleyai.com\/wordpress\/2012\/12\/19\/project-sherlock\/\">Project Sherlock<\/a> and <a href=\"http:\/\/haleyai.com\/wordpress\/wp-content\/uploads\/2013\/06\/Acquiring-Rich-Logical-Knowledge-from-Text-SemTech2013.pptx\">Acquiring Rich Logical Knowledge From Text \u2013 SemTech 2013<\/a><\/p>\n<p><a href=\"#_ftnref7\"><sup><\/sup><sup>[7]<\/sup><\/a> <strong><a href=\"http:\/\/silk.semwebcentral.org\/\">The SILK Project: Semantic Inferencing on Large Knowledge<\/a><\/strong><\/p>\n<p><a href=\"#_ftnref8\"><sup><\/sup><sup>[8]<\/sup><\/a> <strong><a title=\"Permalink to It\u2019s hard to reckon nice English\" href=\"http:\/\/haleyai.com\/wordpress\/2015\/10\/22\/its-hard-to-reckon-nice-english\/\">It\u2019s hard to reckon nice English<\/a><\/strong><\/p>\n","protected":false},"excerpt":{"rendered":"<p>In preparing for some natural language generation[1], I came across some work on natural logic[2][3] and reasoning by textual entailment[4] (RTE) by Richard Bergmair in his PhD at Cambridge: Monte Carlo Semantics: Robust Inference and Logical Pattern Processing with Natural Language Text The work he describes overlaps our approach to robust inference from the deep, &hellip; <\/p>\n<p class=\"link-more\"><a href=\"http:\/\/haleyai.com\/wordpress\/2015\/10\/23\/robust-inference-and-slacker-semantics\/\" class=\"more-link\">Continue reading<span class=\"screen-reader-text\"> &#8220;Robust Inference and Slacker Semantics&#8221;<\/span><\/a><\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[16,243],"tags":[712,640,487,643,530,713,188,527,203,714,711,496,231,715,710,548,495,497,317],"class_list":["post-841","post","type-post","status-publish","format-standard","hentry","category-natural-language","category-question-answering","tag-boxer","tag-ccg","tag-erg","tag-groningen-meaning-bank","tag-hpsg","tag-johan-bos","tag-lfg","tag-linguist","tag-microsoft","tag-minimal-recursion-semantics","tag-monte-carlo","tag-mrs","tag-powerset","tag-richard-bergmair","tag-rte","tag-textual-entailment","tag-under-specified-representation","tag-usr","tag-xerox-parc"],"_links":{"self":[{"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/posts\/841","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/comments?post=841"}],"version-history":[{"count":4,"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/posts\/841\/revisions"}],"predecessor-version":[{"id":1547,"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/posts\/841\/revisions\/1547"}],"wp:attachment":[{"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/media?parent=841"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/categories?post=841"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/haleyai.com\/wordpress\/wp-json\/wp\/v2\/tags?post=841"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}