Lecture 14 Context-Free Grammar

Context-Free Grammar 上下文无关语法

Basics of Context-Free Grammars 上下文无关语法基础

• Symbols: 符号

  • Terminal: word such as book 终结符：如单词book
  • Non-terminal: syntactic label such as NP and VP 非终结符：如句法标签NP和VP
  • convention: 规约
    • lowercase for terminals 使用小写表示终结符
    • uppercase for non-terminals 使用大写表示非终结符

• Productions: 产生式：

  • W -> X Y Z
  • Exactly one non-terminal on the LHS 左侧恰好有一个非终结符
  • An ordered list of symbols on the RHS, can be terminals or non-terminals 右侧为符号的有序列表，可以是终结符或非终结符

• Start symbol: S 开始符号：S

• Context-Free: 上下文无关：

  • Production rules depends only on the LHS, not on ancestors, neighbors 产生规则仅依赖于左侧，不依赖于祖先节点，邻居节点
    • Analogous to Markov chain 类似于马尔可夫链
    • Behaviour at each steps depends only on the current state 每一步的行为仅依赖于当前状态
  • Context-Free languages more general than regular languages. Allows recursive nesting 上下文无关语言比正规语言更通用。允许递归嵌套

CFG Parsing CFG解析

• Given production rules: E.g. 给定产生规则：例如

  • S -> a S b
  • S -> a b

• And a string: aaabbb 以及一个字符串：aaabbb

• Produce a valid parse tree: 生成有效的解析树：

  

• If English can be represented with CFG: 如果英语可以用CFG表示：

  • First develop the production rules 首先开发产生规则
  • Can then build a parser to automatically judge whether a sentence is grammatical 然后可以构建一个解析器来自动判断一个句子是否符合语法

• CFG strike a good balance: CFG取得了良好的平衡：

  • CFG covers most syntactic patterns CFG覆盖了大多数语法模式
  • CFG parsing in computational efficient CFG解析在计算上是高效的

Constituents 成分

Syntactic Constituents 句法成分

• Sentences are broken into constituents 句子被分解成各个成分

  • Word sequence that function as a coherent unit for linguistic analysis 作为语言分析的连贯单元的词序列
  • Helps build CFG production rules 帮助构建CFG产生规则

• Constituents have certain key properties: 成分有一些关键属性：

  • movement: Constituents can be moved around sentences: 移位：成分可以在句子中移动位置：
    • Abigail gave [her brother] [a fish]
    • Abigail gave [a fish] to [her brother]
    • Contrast: [gave her] and [brother a]
  • substitution: Constituents can be substituted by other phrases of the same type: 替换：成分可以被同类型的其他短语替换：
    • Max thanked [his older sister]
    • Max thanked [her]
    • Contrast: [Max thanked] and [thanked his]
  • coordination: Constituents can be conjoined with other coordinators like and and or" 连词：成分可以与and和or等连词结合：
    • [Abigail] and [her young brother] brought a fish
    • Abigail [bought a fish] and [gave it to Max]
    • Abigail [bought] and [greedily ate] a fish

Constituents and Phrases 成分和短语

• Once identify constituents, use phrases to describe them 一旦确定了成分，使用短语来描述它们

• Phrases are determined by their head word: 短语由其头词确定：

  • Noun phrase 名词短语: her younger brother
  • Verb phrase 动词短语: greedily ate it

Example: A Simple CFG for English and generating sentences 示例：英语的简单CFG和生成句子

• 终结符Terminal Symbols: rat, the, ate, cheese

• 非终结符 Non-terminal symbols: S, NP, VP, DT, VBD, NN

• Productions: 产生式：

  • S -> NP VP
  • NP -> DT NN
  • VP -> VBD NP
  • DT -> the
  • NN -> rat
  • NN -> cheese
  • VBD -> ate

• Generating Sentences with CFGs: 使用CFG生成句子：

  

CFG Trees CFG树

• Generation corresponds to a syntactic tree 生成对应于句法树

• Non-terminals are internal nodes 非终结符是内部节点

• Terminals are leaves 终结符是叶节点

• CFG parsing is the reverse process CFG解析是逆向过程

• E.g.:

  

CYK Algorithm CYK算法

CYK Algorithm CYK算法

• Bottom-up parsing 自底向上解析

• Tests whether a string is valid given a CFG, without enumerating all possible parses 检查一个字符串是否符合给定的CFG，而无需列举所有可能的解析

• Core idea: Form small constituents first, and merge them into larger constituents 核心思想：首先形成小的成分，然后将它们合并成更大的成分

• Requirement: CFGs must be in Chomsky Normal Forms 要求：CFG必须在Chomsky正则形式中

Convert to Chomsky Normal Form 转化为乔姆斯基正则形式

• Change grammar so all rules of form: 改变语法，使所有规则满足以下形式：

  • A -> B C: Non-terminal LHS to two non-terminals RHS A -> B C：非终结符的左侧转化为两个非终结符的右侧
  • A -> a: Non-terminal LHS to one terminal RHS A -> a：非终结符的左侧转化为一个终结符的右侧

• To meet requirements 为了满足要求

  • convert rules of form A -> B c into: 将形式A -> B c的规则转化为：

    • A -> B X and X -> c

  • convert rules of form A -> B C D into: 将形式A -> B C D的规则转化为：

    • A -> B Y and Y -> C D

• CNF disallows unary rules like A -> B to avoid infinite loops CNF不允许A -> B这样的一元规则以避免无限循环

  • Replace RHS non-terminal with its productions: 使用其产生式替换右侧的非终结符：
    • A -> B, B -> cat, B -> dog -> A -> cat, A -> dog

The CYK Parsing Algorithm CYK解析算法

• Convert grammar to Chomsky Normal Form 将语法转化为乔姆斯基正则形式

• Fill in a parse table, left to right, bottom to top 填充解析表格，从左到右，从下到上

• Use table to derive parse 使用表格派生解析

• S in top right corner of table -> success S在表格的右上角 -> 成功

• Convert result back to original grammar 将结果转化回原来的语法

• E.g.

  

• Retrieving the Parses 检索解析

  • S in the top-right corner of parse table indicates success 解析表的右上角的S表示成功

  • To get parses, follow pointer back for each match: 获取解析，回溯每个匹配的指针：

    

* If multiple solutions are available, all of the trees are valid: 如果有多个解决方案，所有的树都是有效的 》 

• Pseudo Code:

function CYK-Parse(words, grammar) returns table
    for j <- from 1 to LENGTH(words) do
        for all {A | A -> words[j] ∈ grammar}
            table[j-1, j] <- table[j-1, j] ∪ A
        for i <- from j-2 down to 0 do
            for k <- i + 1 to j - 1 do
                for all {A | A -> BC ∈ grammar and B ∈ table[i, k] and C ∈ table[k, j]
                    table[i, j] <- table[i, j] ∪ A
                    
    return table

Representing English with CFGs

From Toy Grammars to Real Grammars

• Toy grammars with handful productions good for demonstration or extremely limited domains

• For real texts, we need real grammars

• Many thousands of production rules

Key Constituents in Penn Treebank

• Sentence S
• Noun phrase NP
• Verb phrase VP
• Prepositional phrase PP
• Adjective phrase AdjP
• Adverbial phrase AdvP
• Subordinate clause SBAR
• E.g.

  

Basic English Sentence Structures

• Declarative sentences S -> NP VP

  • The rat ate the cheese

• Imperative sentences S -> VP

  • Eat the cheese

• Yes/no questions S -> VB NP VP

  • Did the rat eat the cheese?

• Wh-subject questions S -> WH VP

  • Who ate the cheese

• Wh-object questions S -> WH VB NP VP

  • What did the rat eat?

English Noun Phrases

• Pre-modifiers:

  • DT, CD, ADJP, NNP, NN
  • E.g.: the two very best Philly cheese steaks

• Post-modifiers:

  • PP, VP, SBAR
  • E.g.: A delivery from Bob coming today that I don't want to miss

English Verb Phrases

• Auxiliaries

  • MD, AdvP, VB, TO
  • E.g.: should really have tried to wait

• Arguments and adjuncts

  • NP, PP, SBAR, VP, AdvP
  • E.g.: told him yesterday that I was ready

Other Constituents

• Prepositional phrase:

  • PP -> IN NP
  • E.g.: in the house

• Adjective phrase:

  • AdjP -> (AdvP) JJ
  • E.g.: really nice

• Adverb phrase:

  • AdvP -> (AdvP) RB
  • not too well

• Subordinate clause

  • SBAR -> (IN) S
  • E.g.: since I came here

• Coordination

  • NP -> NP CC NP; VP -> VP CC VP
  • E.g.: Jack and Jill

• Complex sentences

  • S -> S SBAR; S -> SBAR S
  • E.g.: if he goes, I'll go