英文句子的语义分析 读书报告 & 初步结果 刘庆霞 20170309
语义分析 语义分析 产生意义表示,并将其指派给输入语句的过程 句法驱动的语义分析 组合性原则 一个句子的意义可以由它的几个部分的意义组合而成
意义表示 (Meaning Representation) The meaning of linguistic utterances can be captured in formal structures, which we will call meaning representations. 意义表示的计算要求 Verifiability Unambiguous one representation -> one meaning Canonical Form one meaning -> one representation Inference and Variables Expresiveness
意义表示 (Meaning Representation) 一阶谓词演算(FOPC) 项(term) 常量(constant) 连接词(connective) 量词(quantifier) 变量(variable) 谓词(Predicate) 函数(function) Stanford: DCS; Edingburg:CCG Texas Unv. at Austin: lambda calculus (德克萨斯大学奥斯汀分校)
句法驱动的语义分析 为上下文无关语法规则 扩充 语义附着(semantic attachment) S -> NP VP {VP.sem(NP.sem)} VP -> Verb NP {Verb.sem(NP.sem)} Verb -> serves {λx∃e, y Isa(e, Serving) ∧Server(e,y) ∧Served(e,x)}
英语片段的语义 名词短语 形容词短语 动词短语 介词短语 复合名词、所属名词短语 名词前形容词、谓语性形容词 不定式动词短语 Who told Harry to go to Maharani. 介词短语 修饰名词、修饰动词短语、作为动词论元
名词短语 复合名词 Summer light schedule 所属名词短语 Atlanta’s airport 含明前形容词的NP I don’t mind a cheap restaurant 交集语义 λx Isa(x, Restaurant) ∧ Isa (x, Cheap) small elephant,fake gun 仅指出存在抽象关系 λx Isa(x, Restaurant) ∧ AM (x, Cheap)
介词短语 修饰名词 修饰动词 作为动词的论元 a restaurant on Pearl ∃x Isa(x, Resuaurant) ∧On(x, Pearl) 修饰动词 ate dinner in a hurry ∃y Isa(e, Eating) ∧Eater(e, y) ∧Eaten(e, Dinner) ∧In(e, < ∃h Hurry(h)>) 作为动词的论元 I need to go from Boston to Dallas. 问句中,修饰动词的情况比较少见; <>复杂项:因为FOPC(一阶谓词演算)不允许将函数作为论元; from Boston=Boston 名词短语的语义直接复制给介词短短语,然后由动词短语将该意义表示指派到恰当的角色
句法驱动的语义分析 缺点 视野有限 上下文独立 推理无关
Pattern-based Semantic Analysis Train sentence -> pattern What is the revenue of IBM? what be the X of E ? assing semantic representation to frequent patterns (?what, the X of, E) generate semantic representation for unfreq patterns what be the X of X , X ? What is the total population of Melbourne, Florida? what be the X of the X of X ? What is the official website of Tom Hanks? what be the X of the X where X 's X V ? What is the name of the school where Obama's wife studied? Test pattern match and alignment generate semantic representation (pattern, semantic) What is the name of the school where Obama's wife studied?
Step1: Pattern Generation Item Types E : named entities V : normal verbs (e.g. VBX) X : meaningful words (e.g. NN, PRP, CD, JJ, RB) Struct: other words except for E,V,X (e.g., WHX, IN, TO, CC) what, for, in, and, or frequent words many, there, have, be, do
Frequent Patterns Free917 458 patterns Pattern num sentence who V E 42 who designed the iphone what be the X of E 40 what are the deities of hinduism how many X do E have 19 how many floors does the white house have when be E V 18 when was wells fargo founded when be E X 17 when was the airspeed oxford first flown what be E 's X 15 what are bmw's manufacturing plants what X do E X 14 what conferences does google sponsor when be X V when was savealot founded what X do E V 12 what lines does the london overground operate where be E X 11 where was jerry seinfeld born how many X V in E 10 how many teams participate in the uefa
Frequent Patterns QALD679 441 patterns Pattern num sentence give I all X . 17 Give me all school types. who be the X of E ? Who is the author of WikiLeaks? what be the X of E ? 14 What is the revenue of IBM? which X have the X ? 9 Which country has the most official languages? who V E ? Who created Goofy? who be the X of X ? 8 Who was the wife of President Lincoln? who V X ? Who created English Wikipedia? which X have X than X ? Which caves have more than 3 entrances? how many X do E have ? How many employees does IBM have? what be the X of X ? 7 What is the official website of Tom Hanks?
Complex Patterns who be X under the X who V the X of X against X during X ? Who was vice president under the president who approved the use of atomic weapons against Japan during World War II? what X V [X and X for X V X from X] what movies won ascap film and television music awards for most performed songs from motion pictures mothion pictures 电影 complex NP: most performed songs from motion pictures(最佳电影歌曲奖)
Next Syntactic Structure Semantic Annotation 归纳,不能确保具有相同的语义; 需要语义标注信息,总结相似转换模式的句子;
Thank you