基于统计语言模型的依存句法分析技术研究

Dependency

Parsing

4星 · 超过85%的资源需积分: 9 57 浏览量更新于2024-07-31 1 收藏 1.06MB PDF 举报

身份认证购VIP最低享 7 折!

领优惠券(最高得80元）

Dependency Parsing Dependency Parsing是自然语言处理（NLP）和计算语言学（CL）领域中的一个重要主题。它是对句子中词语之间的语法关系的分析和表征。Dependency Parsing的主要目的是将句子中的词语组织成一个树形结构，从而表征词语之间的语法关系。在 Dependency Parsing 中，句子中的词语被分为头词（head）和修饰词（dependent）。头词是句子中中心词语，而修饰词是与头词相关的词语。例如，在句子"The dog chased the cat"中，"dog"是头词，而"chased"和"cat"是修饰词。Dependency Parsing 的目标是确定头词和修饰词之间的语法关系，例如，"chased"是"dog"的动词，而"cat"是"chased"的宾语。 Dependency Parsing 技术有很多应用，例如： * 信息检索（Information Retrieval）：Dependency Parsing 可以用于改善搜索引擎的查询结果，例如，通过分析句子结构，搜索引擎可以更好地理解用户的查询意图。 * 机器翻译（Machine Translation）：Dependency Parsing 可以用于改善机器翻译的质量，例如，通过分析句子结构，机器翻译系统可以更好地理解源语言和目标语言之间的语法关系。 * 文本分类（Text Classification）：Dependency Parsing 可以用于改善文本分类的准确性，例如，通过分析句子结构，文本分类系统可以更好地理解文本的语义。本书"Dependency Parsing"由 Sandra Kübler、Ryan McDonald 和 Joakim Nivre 合著，出版于 2009 年，是 Synthesis Lectures on Human Language Technologies 系列丛书的一部分。这本书提供了 Dependency Parsing 的详细介绍，包括其基本概念、算法和应用。在本书中，作者们首先介绍了 Dependency Parsing 的基本概念，例如，词语之间的语法关系、头词和修饰词的定义等。然后，作者们介绍了各种 Dependency Parsing 算法，例如， Maximum Entropy 算法、 Support Vector Machines 算法等。最后，作者们讨论了 Dependency Parsing 在信息检索、机器翻译和文本分类等领域的应用。 Dependency Parsing 是自然语言处理和计算语言学领域中的一个重要主题，它有很多应用和潜力。本书"Dependency Parsing"提供了该领域的详细介绍和实践指南，对于从事自然语言处理和计算语言学领域的研究人员和开发人员非常有价值。

资源详情

资源推荐

4 CHAPTER 1. INTRODUCTION

4. H selects D and determines whether D is obligatory or optional.

5. The form of D depends on H (agreement or government).

6. The linear position of D is speciﬁed with reference to H .

It is clear that this list contains a mix of different criteria, some syntactic and some semantic, and one

may ask whether there is a single coherent notion of dependency corresponding to all the different

criteria. Some theorists therefore posit the existence of several layers of dependency structure, such

as morphology, syntax and semantics, or surface syntax and deep syntax. Others have pointed out

the need to have different criteria for different kinds of syntactic constructions, in particular for

endocentric and exocentric constructions.

In ﬁgure 1.1, the attribute relation (ATT) holding between the noun markets and the adjective

ﬁnancial is an endocentric construction, where the head can replace the whole without disrupting

the syntactic structure:

Economic news had little effect on [ﬁnancial] markets.

Endocentric constructions may satisfy all of the criteria listed above, although number 4 is usually

considered less relevant, since dependents in endocentric constructions are taken to be optional and

not selected by their heads.By contrast,the prepositional complement relation (PC) holding between

the preposition on and the noun markets is an exocentric construction, where the head cannot readily

replace the whole:

Economic news had little effect on [markets].

Exocentric constructions, by their deﬁnition, fail on criterion number 1, at least with respect to sub-

stitutability of the head for the whole, but may satisfy the remaining criteria. Considering the rest

of the relations exempliﬁed in ﬁgure 1.1, the subject and object relations (SBJ, OBJ) are clearly exo-

centric, and the attribute relation from the noun news to the adjective Economic clearly endocentric,

while the remaining attribute relations (effect → little, effect → on) have a less clear status.

The distinction between endocentric and exocentric constructions is also related to the dis-

tinction between head-complement and head-modiﬁer (or head-adjunct) relations found in many con-

temporary syntactic theories, since head-complement relations are exocentric while head-modiﬁer

relations are endocentric. The distinction between complements and modiﬁers is often deﬁned in

terms of valency, which is a central notion in the theoretical tradition of dependency grammar.

Although the exact characterization of this notion differs from one theoretical framework to the

other, valency is usually related to the semantic predicate-argument structure associated with certain

classes of lexemes, in particular verbs but sometimes also nouns and adjectives. The idea is that the

verb imposes requirements on its syntactic dependents that reﬂect its interpretation as a semantic

predicate. Dependents that correspond to arguments of the predicate can be obligatory or optional

in surface syntax but can only occur once with each predicate. By contrast, dependents that do not

correspond to arguments can have more than one occurrence with a single predicate and tend to be

1.1. DEPENDENCY GRAMMAR 5

Figure 1.3: Two analyses of coordination in dependency grammar.

optional.The valency frame of the verb is normally taken to include argument dependents, but some

theories also allow obligatory non-arguments to be included. Returning to ﬁgure 1.1, the subject

and the object would normally be treated as valency-bound dependents of the verb had, while the

adjectival modiﬁers of the nouns news and markets would be considered valency-free. The preposi-

tional modiﬁcation of the noun effect may or may not be treated as valency-bound, depending on

whether the entity undergoing the effect is supposed to be an argument of the noun effect or not.

Another term that is sometimes used in connection with valency constraints is arity, which primarily

refers to the number of arguments that a predicate takes (without distinguishing the types of these

arguments).

While most head-complement and head-modiﬁer structures have a straightforward analysis

in dependency grammar, there are also constructions that have a more unclear status. This group

includes constructions that involve grammatical function words, such as articles, complementizers

and auxiliary verbs, but also structures involving prepositional phrases. For these constructions, there

is nogeneralconsensus in the tradition of dependency grammar astowhether they should be analyzed

as dependency relations at all and, if so, what should be regarded as the head and what should be

regarded as the dependent. For example, some theories regard auxiliary verbs as heads taking lexical

verbs as dependents; other theories make the opposite assumption; and yet other theories assume

that verb chains are connected by relations that are not dependencies in the usual sense.

Another kind of construction that is problematic for dependency grammar (as for most the-

oretical traditions) is coordination. According to the structuralist tradition, coordination is an en-

docentric construction, since it contains not only one but several heads that can replace the whole

construction syntactically. However, this raises the question of whether coordination can be analyzed

in terms of binary relations holding between a head and a dependent. Consider the following simple

examples:

They operate ships and banks.

She bought and ate an apple.

In the ﬁrst example, it seems clear that the phrase ships and banks functions as a direct object of the

verb operate, but it is not immediately clear how this phrase can be given an internal analysis that is

compatible with the basic assumptions of dependency grammar, since the two nouns ships and banks

seem to be equally good candidates for being heads. Similarly, in the second example, the noun apple

6 CHAPTER 1. INTRODUCTION

is the object of the coordinated verb group bought and ate, where in some sense both verbs function

as the head of the noun. The most popular treatments of coordination in dependency grammar are

illustrated for the ﬁrst example in ﬁgure 1.3, where the analysis to the left treats the conjunction as

the head, an analysis that may be motivated on semantic grounds, while the analysis on the right

treats the conjunction as the head only of the second conjunct and analyzes the conjunction as a

dependent of the ﬁrst conjunct. The arguments for the latter analysis are essentially the same as the

arguments for an asymmetric right-branching analysis in phrase structure grammar.

To sum up, the theoretical tradition of dependency grammar is united by the assumption that

syntactic structure essentially consists of dependency relations between words. Moreover, there is

a core of syntactic constructions for which the analysis given by different frameworks agree in all

important respects, notably predicate-argument and head-modiﬁer constructions. However, there

are also constructions for which there is no clear consensus, such as verb groups and coordination.

Finally, it is worth pointing out that the inventory of dependency types used to classify dependency

relations vary from one framework to the other. Besides traditional grammatical functions (such

as predicate, subject, and object), semantic roles (such as agent, patient, and goal) are commonly

used, especially in representations of deep syntax and semantics. Another dimension of variation

is the number of representational levels, or strata, assumed in different theories. Although we will

concentrate in this book on mono-stratal representations, using a single dependency structure for

syntactic analysis, many theoretical frameworks make use of multi-stratal representations,often with

different levels for syntax and semantics.

1.2 DEPENDENCY PARSING

Having introduced the basic notions of dependency grammar, we will now turn to the problem

of dependency parsing , that is, the task of automatically analyzing the dependency structure of a

given input sentence. Throughout this book we will consider a number of different methods for

solving this problem, some based on inductive machine learning from large sets of syntactically

annotated sentences, others based on formal grammars deﬁning permissible dependency structures.

Common to all of these methods is that they do not make any speciﬁc assumptions about the kind

of dependency types used, be they grammatical functions or semantic roles, nor about the speciﬁc

analysis of different linguistic constructions, such as verb groups or coordination.

All that is assumed is that the task of the parser is to produce a labeled dependency structure

of the kind depicted in ﬁgure 1.1, where the words of the sentence (including the artiﬁcial word

root) are connected by typed dependency relations. This will be made more precise in chapter 2,

where we deﬁne dependency structures as labeled directed graphs – called dependency graphs – and

discuss a number of formal properties of these structures. But for the time being we can deﬁne the

parsing problem as that of mapping an input sentence S, consisting of words w

...w

(where

= root), to its dependency graph G. In the remainder of this chapter, we will give an overview

of the different approaches to this problem that are covered in the book.

1.2. DEPENDENCY PARSING 7

Broadly speaking, these approaches can be divided into two classes, which we will call data-

driven and grammar-based, respectively. An approach is data-driven if it makes essential use of

machine learning from linguistic data in order to parse new sentences. An approach is grammar-based

if it relies on a formal grammar, deﬁning a formal language, so that it makes sense to ask whether a

given input sentence is in the language deﬁned by the grammar or not. It is important to note that

these categorizations are orthogonal, since it is possible for a parsing method to make essential use

of machine learning and use a formal grammar, hence to be both data-driven and grammar-based.

However, most of the methods that we cover fall into one of these classes only.

The major part of the book, chapters 3–4 to be exact, is devoted to data-driven methods for

dependency parsing, which have attracted the most attention in recent years. We focus on supervised

methods, that is, methods presupposing that the sentences used as input to machine learning have

been annotated with their correct dependency structures. In supervised dependency parsing, there

are two different problems that need to be solved computationally. The ﬁrst is the learning problem,

which is the task of learning a parsing model from a representative sample of sentences and their

dependency structures. The second is the parsing problem, which is the task of applying the learned

model to the analysis of a new sentence.

We can represent this as follows:

• Learning: Given a training set

D of sentences (annotated with dependency graphs), induce a

parsing model M that can be used to parse new sentences.

• Parsing: Given a parsing model M and a sentence S, derive the optimal dependency graph G

for S according to M.

Data-driven approaches differ in the type of parsing model adopted, the algorithms used to learn

the model from data, and the algorithms used to parse new sentences with the model. In this book,

we focus on two classes of data-driven methods, which we call transition-based and graph-based,

respectively. These classes contain most of the methods for data-driven dependency parsing that

have been proposed in recent years.

Transition-based methods start by deﬁning a transition system, or state machine, for mapping

a sentence to its dependency graph. The learning problem is to induce a model for predicting the

next state transition, given the transition history, and the parsing problem is to construct the optimal

transition sequence for the input sentence,given the induced model.This is sometimes referred to as

shift-reduce dependency parsing, since the overall approach is inspired by deterministic shift-reduce

parsing for context-free grammars. Transition-based approaches are treated in chapter 3.

Graph-based methods instead deﬁne a space of candidate dependency graphs for a sentence.

The learning problem is to induce a model for assigning scores to the candidate dependency graphs

for a sentence, and the parsing problem is to ﬁnd the highest-scoring dependency graph for the input

sentence, given the induced model. This is often called maximum spanning tree parsing, since the

problem of ﬁnding the highest-scoring dependency graph is equivalent, under certain assumptions,

The parsing problem is sometimes referred to as the inference problem or decoding problem, which are the general terms used in

machine learning for the application of a learned model to new data.

8 CHAPTER 1. INTRODUCTION

to the problem of ﬁnding a maximum spanning tree in a dense graph. Graph-based approaches are

treated in chapter 4.

Most data-driven approaches,whether transition-basedorgraph-based,assume that any input

string is a valid sentence and that the task of the parser is to return the most plausible dependency

structure for the input, no matter how unlikely it may be. Grammar-based approaches, by contrast,

make use of a formal grammar that only accepts a subset of all possible input strings. Given our

previous characterization of the parsing problem,we may say that this formal grammar is an essential

component of the model M used to parse new sentences. However, the grammar itself may be hand-

crafted or learned from linguistic data, which means that a grammar-based model may or may not

be data-driven as well. In chapter 5, we discuss selected grammar-based methods for dependency

parsing, dividing them into two classes, which we call context-free and constraint-based, respectively.

Context-free dependency parsing exploits a mapping from dependency structures to context-

free phrase structure representations and reuses parsing algorithms originally developed for context-

free grammar.This includes chart parsing algorithms, which are also used in graph-based parsing, as

well as shift-reduce type algorithms,which are closely related to the methods used in transition-based

parsing.

Constraint-based dependency parsing views parsing as a constraint satisfaction problem. A

grammar is deﬁned as a set of constraints on well-formed dependency graphs, and the parsing

problem amounts to ﬁnding a dependency graph for a sentence that satisﬁes all the constraints of

the grammar. Some approaches allow soft, weighted constraints and score dependency graphs by a

combination of the weights of constraints violated by that graph. Parsing then becomes the problem

of ﬁnding the dependency graph for a sentence that has the best score, which is essentially the same

formulation as in graph-based parsing.

We can sum up our coverage of dependency parsing methods as follows:

• Data-driven dependency parsing

– Transition-based dependency parsing (chapter 3)

– Graph-based dependency parsing (chapter 4)

• Grammar-based parsing (chapter 5)

– Context-free dependency parsing

– Constraint-based dependency parsing

In chapter 6, we discuss issues concerning evaluation, both the evaluation of dependency parsers and

the use of dependencies as a basis for cross-framework evaluation, and in chapter 7, we compare the

approaches treated in earlier chapters, pointing out similarities and differences between methods, as

well as complementary strengths and weaknesses. We conclude the book with some reﬂections on

current trends and future prospects of dependency parsing in chapter 8.

剩余126页未读，继续阅读

lrjnlp

粉丝: 4
资源: 21

基于统计语言模型的依存句法分析技术研究

袁蕴哲__Bidirectional Transition-Based Dependency Parsing1

A Pipeline Framework for Dependency Parsing

dependency parsing

语义角色标注可以设置那些辅助任务

依存句法分析属于神经网络么

Python 特征抽取

Python依存句法分析

聊天机器人通常使用哪些自然语言处理技术？

怎么使用python理清文本中的逻辑关系

自然语言处理核心算法

如何用自然语言处理（NLP）技术来解析文本

学习并介绍一个自己感兴趣的第三方库， 包括功能、学习过程和学习成果（若有程序和运行截图，可放于附录中）。

这是什么意思：entity_dict_in_file = defaultdict(lambda: defaultdict(list))

对这十篇文档使用textrank方法抽取top-20的关键词，并给我数据、代码和结果

stanford-corenlp

Problem parsing dependency 21 of dbus.cpp.bin:arm64=5,0,1-4

quarkus restclient RESTEASY003145: Unable to find a MessageBodyReader of content-type applicati

最新资源

学习并介绍一个自己感兴趣的第三方库，包括功能、学习过程和学习成果（若有程序和运行截图，可放于附录中）。