语音与语言处理:自然语言处理、计算语言学和语音识别导论:第2版:英文(全面系统地介绍了计算机自然语言处理)

.16 language and complexity 　　 563
iv semantics and pragmatics
17 the representation ofmeaning　 579
18 computational semantics 　　 617
19 lexical semantics　 645
20 computational lexical semantics　 671
21 computational discourse 　　 715
v applications
22 information extraction 　　 759
23 question answering and summarization 799
24 dialogue and conversational agents 847
25 machine translation 　　 895
bibliography 945
author index 995
subject index 1007
contents
foreword 23
preface 25
about the authors 31
1 introduction 35
1.1 knowledge in speech and language processing 　 36
1.2 ambiguity 38
1.3 models andalgorithms 39
1.4 language, thought, and understanding 　　 40
1.5 thestate of theart 42
1.6 somebriefhistory 43
1.6.1 foundational insights: 1940s and 1950s 　 43
1.6.2 the two camps: 1957–1970 　　 44
1.6.3 four paradigms: 1970–1983 　　 45
1.6.4 empiricism and finite-state models redux: 1983–1993 　 46
1.6.5 the field comes together: 1994–1999　 46
1.6.6 the rise of machine learning: 2000–2008 　 46
1.6.7 on multiple discoveries 　 47
1.6.8 a final brief note on psychology 　　 48
1.7 summary 　 48
bibliographical and historical notes 　 49
i words
2 regular expressions and automata　 51
2.1 regularexpressions 　 51
2.1.1 basic regular expression patterns 　　 52
2.1.2 disjunction, grouping, and precedence　 55
2.1.3 asimpleexample　 56
2.1.4 a more complex example　 57
2.1.5 advancedoperators 　 58
2.1.6 regular expression substitution, memory, and eliza 　 59
2.2 finite-stateautomata 　 60
2.2.1 use of an fsa to recognize sheeptalk 　 61
2.2.2 formal languages　 64
2.2.3 another example 　 65
2.2.4 non-deterministic fsas . 66
2.2.5 use of an nfsa to accept strings 　 67
2.2.6 recognition as search 69
2.2.7 relation of deterministic and non-deterministic automata 　 72
foreword 　 23
preface 　 25
about the authors　 31
1 introduction 　 35
1.1 knowledge in speech and language processing　 36
1.2 ambiguity 　 38
1.3 models andalgorithms 　 39
1.4 language, thought, and understanding 　　　40
1.5 thestate of theart . 42
1.6 somebriefhistory . 43
1.6.1 foundational insights: 1940s and 1950s 43
1.6.2 the two camps: 1957–1970 　　 44
1.6.3 four paradigms: 1970–1983 　　 45
1.6.4 empiricism and finite-state models redux: 1983–1993 46
1.6.5 the field comes together: 1994–1999 46
1.6.6 the rise of machine learning: 2000–2008 46
1.6.7 on multiple discoveries 47
1.6.8 a final brief note on psychology 　　 48
1.7 summary 　 48
bibliographical and historical notes 49
i words
2 regular expressions and automata 51
2.1 regularexpressions 51
2.1.1 basic regular expression patterns 　　 52
2.1.2 disjunction, grouping, and precedence　 55
2.1.3 asimpleexample　 56
2.1.4 a more complex example 　 57
2.1.5 advancedoperators 　 58
2.1.6 regular expression substitution, memory, and eliza　 59
2.2 finite-stateautomata　 60
2.2.1 use of an fsa to recognize sheeptalk　 61
2.2.2 formal languages　 64
2.2.3 another example 　 65
2.2.4 non-deterministic fsas 　 66
2.2.5 use of an nfsa to accept strings 　　 67
2.2.6 recognition as search　 69
2.2.7 relation of deterministic and non-deterministic automata　 72
2.3 regular languages and fsas　 72
2.4 summary 　 75
bibliographical and historical notes 76
exercises 76
3 words and transducers 79
3.1 survey of (mostly) english morphology 　 81
3.1.1 in?ectional morphology 　 82
3.1.2 derivational morphology　 84
3.1.3 cliticization 　 85
3.1.4 non-concatenative morphology 　　 85
3.1.5 agreement 　 86
3.2 finite-state morphological parsing　 86
3.3 construction of a finite-state lexicon 　　 88
3.4 finite-statetransducers 　 91
3.4.1 sequential transducers and determinism 　 93
3.5 fsts for morphological parsing 　 94
3.6 transducers and orthographic rules 　　 96
3.7 the combination of an fst lexicon and rules 　 99
3.8 lexicon-free fsts: the porter stemmer 　　 102
3.9 word and sentence tokenization　 102
3.9.1 segmentation in chinese　 104
3.10 detection and correction of spelling errors 　 106
3.11 minimumeditdistance 　 107
3.12 human morphological processing 　 111
3.13 summary 　 113
bibliographical and historical notes 　 114
exercises 115
4 n-grams 　 117
4.1 wordcounting incorpora　 119
4.2 simple (unsmoothed) n-grams　 120
4.3 training andtestsets 　 125
4.3.1 n-gram sensitivity to the training corpus　 126
4.3.2 unknown words: open versus closed vocabulary tasks 　 129
4.4 evaluating n-grams: perplexity 　 129
4.5 smoothing 　 131
4.5.1 laplacesmoothing 　 132
4.5.2 good-turing discounting　 135
4.5.3 some advanced issues in good-turing estimation 　 136
4.6 interpolation 　 138
4.7 backoff 　 139
4.7.1 advanced: details of computing katz backoff α and p 141
4.8 practical issues: toolkits and data formats 　　 142
4.9 advanced issues in language modeling 　　 143
4.9.1 advanced smoothing methods: kneser-ney smoothing 　 143
4.9.2 class-based n-grams　 145
4.9.3 language model adaptation and web use　 146
4.9.4 using longer-distance information: a brief summary 　 146
4.10 advanced: information theory background 　　148
4.10.1 cross-entropy for comparing models 　　 150
4.11 advanced: the entropy of english and entropy rate constancy 152
4.12 summary 　 153
bibliographical and historical notes 154
exercises 155
5 part-of-speech tagging 　 157
5.1 (mostly) english word classes　 158
5.2 tagsets forenglish 　 164
5.3 part-of-speech tagging 　 167
5.4 rule-based part-of-speech tagging 　169
5.5 hmm part-of-speech tagging　 173
5.5.1 computing the most likely tag sequence: an example　 176
5.5.2 formalizing hidden markov model taggers　 178
5.5.3 using the viterbi algorithm for hmm tagging 　 179
5.5.4 extending the hmm algorithm to trigrams 　 183
5.6 transformation-based tagging 　 185
5.6.1 how tbl rules are applied 　　 186
5.6.2 how tbl rules are learned 　　 186
5.7 evaluation and error analysis 　 187
5.7.1 erroranalysis　 190
5.8 advanced issues in part-of-speech tagging 　　 191
5.8.1 practical issues: tag indeterminacy and tokenization 　 191
5.8.2 unknown words . 192
5.8.3 part-of-speech tagging for other languages　 194
5.8.4 tagger combination 197
5.9 advanced: the noisy channel model for spelling 　 197
5.9.1 contextual spelling error correction 　　 201
5.10 summary 　 202
bibliographical and historical notes 203
exercises 205
6 hidden markov and maximum entropy models 207
6.1 markovchains 　 208
6.2 thehiddenmarkovmodel 　 210
6.3 likelihood computation: the forward algorithm 　 213
6.4 decoding: the viterbi algorithm　 218
6.5 hmm training: the forward-backward algorithm 　 220
6.6 maximum entropy models: background 　 227
6.6.1 linearregression 　 228
6.6.2 logistic regression 231
6.6.3 logistic regression: classi?cation 　　233
6.6.4 advanced: learning in logistic regression 　 234
6.7 maximum entropy modeling 　 235
6.7.1 why we call it maximum entropy 　　 239
6.8 maximum entropy markov models 241
6.8.1 decoding and learning in memms 　　 244
6.9 summary 　 245
bibliographical and historical notes 246
exercises 247
ii speech
7 phonetics 　 249
7.1 speech sounds and phonetic transcription　 250
7.2 articulatory phonetics 　 251
7.2.1 thevocalorgans 　 252
7.2.2 consonants: place of articulation 　　254
7.2.3 consonants: manner of articulation 　　 255
7.2.4 vowels 256
7.2.5 syllables 257
7.3 phonological categories and pronunciation variation 259
7.3.1 phonetic features . 261
7.3.2 predicting phonetic variation 　　 . 262
7.3.3 factors in?uencing phonetic variation 　　 263
7.4 acoustic phonetics and signals 264
7.4.1 waves 　 264
7.4.2 speech sound waves 　 265
7.4.3 frequency and amplitude; pitch and loudness 　 267
7.4.4 interpretation of phones from a waveform　 270
7.4.5 spectra and the frequency domain 　 270
7.4.6 the source-filter model 　　274
7.5 phonetic resources 　 275
7.6 advanced: articulatory and gestural phonology 　 278
7.7 summary 　 279
bibliographical and historical notes　 280
exercises 　 281
8 speech synthesis 　283
8.1 textnormalization 　 285
8.1.1 sentence tokenization 　 285
8.1.2 non-standard words 　 286
8.1.3 homograph disambiguation 　　290
8.2 phonetic analysis 　 291
8.2.1 dictionary lookup 　 291
8.2.2 names 　 292
8.2.3 grapheme-to-phoneme conversion 　　 293
8.3 prosodicanalysis 　 296
8.3.1 prosodicstructure　 296
8.3.2 prosodic prominence 　 297
8.3.3 tune 　 299
8.3.4 more sophisticated models: tobi 　 300
8.3.5 computing duration from prosodic labels 　302
8.3.6 computing f0 from prosodic labels 　　303
8.3.7 final result of text analysis: internal representation　 305
8.4 diphone waveform synthesis 　 306
8.4.1 steps for building a diphone database 306
8.4.2 diphone concatenation and td-psola for prosody 　308
8.5 unit selection (waveform) synthesis　 310
8.6 evaluation 　 314
bibliographical and historical notes 　 315
exercises 　 318
9 automatic speech recognition 　　319
9.1 speech recognition architecture 　 321
9.2 the hidden markov model applied to speech 　 325
9.3 feature extraction: mfcc vectors　 329
9.3.1 preemphasis　 330
9.3.2 windowing 　 330
9.3.3 discrete fourier transform 　 332
9.3.4 mel filter bank and log 　 333
9.3.5 the cepstrum: inverse discrete fourier transform　 334
9.3.6 deltas andenergy　 336
9.3.7 summary:mfcc 　 336
9.4 acoustic likelihood computation　 337
9.4.1 vector quantization 　 337
9.4.2 gaussianpdfs 　 340
9.4.3 probabilities, log-probabilities, and distance functions　 347
9.5 the lexicon and language model 　 348
9.6 search anddecoding 　 348
9.7 embeddedtraining 　 358
9.8 evaluation: word error rate 362
9.9 summary 　 364
bibliographical and historical notes 　 365
exercises 　 367
10 speech recognition: advanced topics　 369
10.1 multipass decoding: n-best lists and lattices 　　 369
10.2 a? (“stack”)decoding　 375
10.3 context-dependent acoustic models: triphones 　 379
10.4 discriminativetraining　 383
10.4.1 maximum mutual information estimation　 384
10.4.2 acoustic models based on posterior classi?ers 385
10.5 modelingvariation 　 386
10.5.1 environmental variation and noise 　　386
10.5.2 speaker variation and speaker adaptation 　 387
10.5.3 pronunciation modeling: variation due to genre 388
10.6 metadata: boundaries, punctuation, and dis?uencies 　 390
10.7 speech recognition by humans　 392
10.8 summary 　 393
bibliographical and historical notes 　 393
exercises 　 394
11 computational phonology 　 395
11.1 finite-state phonology 　 395
11.2 advanced finite-state phonology 　 399
11.2.1 harmony 　 399
11.2.2 templatic morphology　 400
11.3 computational optimality theory 　 401
11.3.1 finite-state transducer models of optimality theory 　 403
11.3.2 stochastic models of optimality theory　 404
11.4 syllabi?cation 　 406
11.5 learning phonology and morphology 　 409
11.5.1 learning phonological rules 　 409
11.5.2 learning morphology 411
11.5.3 learning in optimality theory 　　414
11.6 summary 415
bibliographical and historical notes 　 415
exercises 417
iii syntax
12 formal grammars of english 419
12.1 constituency 420
12.2 context-freegrammars 421
12.2.1 formal de?nition of context-free grammar 425
12.3 some grammar rules for english 　 426
12.3.1 sentence-level constructions 　　426
12.3.2 clauses and sentences 　 428
12.3.3 the noun phrase　 428
12.3.4 agreement 　 432
12.3.5 the verb phrase and subcategorization　 434
12.3.6 auxiliaries 　 436
12.3.7 coordination　 437
12.4 treebanks 438
12.4.1 example: the penn treebank project 　　 438
12.4.2 treebanks as grammars 　 440
12.4.3 treebank searching　 442
12.4.4 heads and head finding 　443
12.5 grammar equivalence and normal form　 446
12.6 finite-state and context-free grammars 　 447
12.7 dependencygrammars 448
12.7.1 the relationship between dependencies and heads 449
12.7.2 categorial grammar 451
12.8 spoken language syntax 　 451
12.8.1 dis?uencies andrepair 　 452
12.8.2 treebanks for spoken language 　　453
12.9 grammars and human processing 　 454
12.10 summary 455
bibliographical and historical notes 　456
exercises 　 458
13 syntactic parsing 　 461
13.1 parsing assearch 　 462
13.1.1 top-downparsing 　 463
13.1.2 bottom-upparsing　 464
13.1.3 comparing top-down and bottom-up parsing 465
13.2 ambiguity 466
13.3 search in the face of ambiguity . 468
13.4 dynamic programming parsing methods 　　 469
13.4.1 ckyparsing 470
13.4.2 the earley algorithm 477
13.4.3 chartparsing 482
13.5 partialparsing . 484
13.5.1 finite-state rule-based chunking 　　 486
13.5.2 machine learning-based approaches to chunking 486
13.5.3 chunking-system evaluations 　　 . 489
13.6 summary 　490
bibliographical and historical notes 　 491
exercises 　 492
14 statistical parsing 　　493
14.1 probabilistic context-free grammars 　 494
14.1.1 pcfgs for disambiguation 　 495
14.1.2 pcfgs for language modeling 　 497
14.2 probabilistic cky parsing of pcfgs 　 498
14.3 ways to learn pcfg rule probabilities 　 501
14.4 problemswithpcfgs 　502
14.4.1 independence assumptions miss structural dependencies betweenrules　 502
14.4.2 lack of sensitivity to lexical dependencies　 503
14.5 improving pcfgs by splitting non-terminals 　 505
14.6 probabilistic lexicalized cfgs 　507
14.6.1 the collins parser 　509
14.6.2 advanced: further details of the collins parser 　 511
14.7 evaluatingparsers 　513
14.8 advanced: discriminative reranking 　 515
14.9 advanced: parser-based language modeling 　　 516
14.10 humanparsing 　517
14.11 summary 　519
bibliographical and historical notes 　 520
exercises 522
15 features and uni?cation　 523
15.1 featurestructures 　524
15.2 uni?cation of feature structures 　 526
15.3 feature structures in the grammar 　531
15.3.1 agreement 　532
15.3.2 headfeatures 　534
15.3.3 subcategorization 　535
15.3.4 long-distance dependencies 　　 540
15.4 implementation of uni?cation　 541
15.4.1 uni?cation data structures 　 541
15.4.2 the uni?cationalgorithm 　 543
15.5 parsing with uni?cation constraints 　 547
15.5.1 integration of uni?cation into an earley parser 　548
15.5.2 uni?cation-based parsing 　 553
15.6 types and inheritance 　 555
15.6.1 advanced: extensions to typing 　 558
15.6.2 other extensions to uni?cation 　 559
15.7 summary 　 559
bibliographical and historical notes　 560
exercises 561
16 language and complexity 　 563
16.1 thechomskyhierarchy 　 564
16.2 ways to tell if a language isn’t regular 　　 566
16.2.1 the pumping lemma 567
16.2.2 proofs that various natural languages are not regular　 569
16.3 is natural language context free?　 571
16.4 complexity and human processing 　 573
16.5 summary 576
bibliographical and historical notes 577
exercises 578
17 the representation of meaning 579
17.1 computational desiderata for representations 　 581
17.1.1 veri?ability 581
17.1.2 unambiguous representations 　582
17.1.3 canonical form 　 583
17.1.4 inference and variables　 584
17.1.5 expressiveness 　585
17.2 model-theoretic semantics　 586
17.3 first-orderlogic 　 589
17.3.1 basic elements of first-order logic 　　 589
17.3.2 variables and quanti?ers . 591
17.3.3 lambdanotation . 593
17.3.4 the semantics of first-order logic 　594
17.3.5 inference 　 595
17.4 event and state representations　 597
17.4.1 representingtime　 600
17.4.2 aspect 　 603
17.5 descriptionlogics 　 606
17.6 embodied and situated approaches to meaning 　 612
17.7 summary 　 614
bibliographical and historical notes 　 614
exercises 616
18 computational semantics　 617
18.1 syntax-driven semantic analysis 　 617
18.2 semantic augmentations to syntactic rules 　 619
18.3 quanti?er scope ambiguity and underspeci?cation 　 626
18.3.1 store and retrieve approaches 　　 626
18.3.2 constraint-based approaches 　　 629
18.4 uni?cation-based approaches to semantic analysis 　 632
18.5 integration of semantics into the earley parser 　 638
18.6 idioms and compositionality 　 639
18.7 summary 　 641
bibliographical and historical notes　 641
exercises 　 643
19 lexical semantics 　645
19.1 wordsenses 　 646
19.2 relations between senses 　 649
19.2.1 synonymy and antonymy 　 649
19.2.2 hyponymy 　 650
19.2.3 semanticfields 　 651
19.3 wordnet: a database of lexical relations 　　 651
19.4 eventparticipants　 653
19.4.1 thematicroles 　 654
19.4.2 diathesis alternations 　656
19.4.3 problems with thematic roles 　　 657
19.4.4 the proposition bank　 658
19.4.5 framenet 　 659
19.4.6 selectional restrictions 　 661
19.5 primitive decomposition 　 663
19.6 advanced: metaphor 665
19.7 summary 　 666
bibliographical and historical notes 　 667
exercises 　 668
20 computational lexical semantics 　 671
20.1 word sense disambiguation: overview 　　 672
20.2 supervised word sense disambiguation 　　 673
20.2.1 feature extraction for supervised learning　 674
20.2.2 naive bayes and decision list classi?ers 　 675
20.3 wsd evaluation, baselines, and ceilings 　 678
20.4 wsd: dictionary and thesaurus methods 　　680
20.4.1 the lesk algorithm 　 680
20.4.2 selectional restrictions and selectional preferences 　 682
20.5 minimally supervised wsd: bootstrapping 　　 684
20.6 word similarity: thesaurus methods 　　 686
20.7 word similarity: distributional methods 　　 692
20.7.1 de?ning a word’s co-occurrence vectors 　 693
20.7.2 measuring association with context 　　695
20.7.3 de?ning similarity between two vectors　 697
20.7.4 evaluating distributional word similarity 　 701
20.8 hyponymy and other word relations 　 701
20.9 semanticrolelabeling 　 704
20.10 advanced: unsupervised sense disambiguation　 708
20.11 summary 709
bibliographical and historical notes 710
exercises 713
21 computational discourse　 715
21.1 discoursesegmentation　 718
21.1.1 unsupervised discourse segmentation 　718
21.1.2 supervised discourse segmentation 　　720
21.1.3 discourse segmentation evaluation 　 722
21.2 textcoherence　 723
21.2.1 rhetorical structure theory 　 724
21.2.2 automatic coherence assignment 　 726
21.3 referenceresolution 　 729
21.4 referencephenomena 　 732
21.4.1 five types of referring expressions 　　 732
21.4.2 information status 　 734
21.5 features for pronominal anaphora resolution 　　 735
21.5.1 features for filtering potential referents　 735
21.5.2 preferences in pronoun interpretation 　 736
21.6 three algorithms for anaphora resolution 　 738
21.6.1 pronominal anaphora baseline: the hobbs algorithm 　 738
21.6.2 a centering algorithm for anaphora resolution 　 740
21.6.3 a log-linear model for pronominal anaphora resolution 　 742
21.6.4 features for pronominal anaphora resolution 　743
21.7 coreference resolution 　 744
21.8 evaluation of coreference resolution 　 746
21.9 advanced: inference-based coherence resolution 　 747
21.10 psycholinguistic studies of reference 　 752
21.11 summary 　753
bibliographical and historical notes 　 754
exercises　 756
v applications
22 information extraction 　 759
22.1 named entity recognition 　 761
22.1.1 ambiguity in named entity recognition 　 763
22.1.2 ner as sequence labeling 　 763
22.1.3 evaluation of named entity recognition　 766
22.1.4 practical ner architectures 　　 768
22.2 relation detection and classi?cation 　　 768
22.2.1 supervised learning approaches to relation analysis 769
22.2.2 lightly supervised approaches to relation analysis . 772
22.2.3 evaluation of relation analysis systems . 776
22.3 temporal and event processing 777
22.3.1 temporal expression recognition 　　 777
22.3.2 temporal normalization 　 780
22.3.3 event detection and analysis 　　 783
22.3.4 timebank 　784
22.4 template filling 　786
22.4.1 statistical approaches to template-filling 　 786
22.4.2 finite-state template-filling systems 　　 788
22.5 advanced: biomedical information extraction 　　 791
22.5.1 biological named entity recognition 　　 792
22.5.2 gene normalization　 793
22.5.3 biological roles and relations 　 794
22.6 summary 　 796
bibliographical and historical notes　 796
exercises 　 797
23 question answering and summarization　 799
23.1 informationretrieval 　 801
23.1.1 the vector space model 　 802
23.1.2 termweighting 　 804
23.1.3 term selection and creation 　 806
23.1.4 evaluation of information-retrieval systems 806
23.1.5 homonymy, polysemy, and synonymy 　 810
23.1.6 ways to improve user queries 　　810
23.2 factoid question answering　 812
23.2.1 question processing 　 813
23.2.2 passageretrieval　 815
23.2.3 answerprocessing　 817
23.2.4 evaluation of factoid answers 　　 821
23.3 summarization 　 821
23.4 single-document summarization 　 824
23.4.1 unsupervised content selection 　　 824
23.4.2 unsupervised summarization based on rhetorical parsing 　 826
23.4.3 supervised content selection 　　 828
23.4.4 sentence simpli?cation 　 829
23.5 multi-document summarization　 830
23.5.1 content selection in multi-document summarization　 831
23.5.2 information ordering in multi-document summarization 　 832
23.6 focused summarization and question answering 　 835
23.7 summarization evaluation 　 839
23.8 summary 　 841
bibliographical and historical notes 　 842
exercises 844
24 dialogue and conversational agents 　847
24.1 properties of human conversations 　849
24.1.1 turns and turn-taking 　849
24.1.2 language as action: speech acts 　　 851
24.1.3 language as joint action: grounding 　 852
24.1.4 conversational structure 　 854
24.1.5 conversational implicature　 855
24.2 basic dialogue systems 　 857
24.2.1 asr component　 857
24.2.2 nlu component 　 858
24.2.3 generation and tts components 　 861
24.2.4 dialogue manager 　 863
24.2.5 dealing with errors: con?rmation and rejection 867
24.3 voicexml 868
24.4 dialogue system design and evaluation 　　 872
24.4.1 designing dialogue systems 　　 872
24.4.2 evaluating dialogue systems 　 872
24.5 information-state and dialogue acts 　 874
24.5.1 using dialogue acts 　 876
24.5.2 interpreting dialogue acts　 877
24.5.3 detecting correction acts　 880
24.5.4 generating dialogue acts: con?rmation and rejection 　881
24.6 markov decision process architecture 　　 882
24.7 advanced: plan-based dialogue agents 　　 886
24.7.1 plan-inferential interpretation and production　 887
24.7.2 the intentional structure of dialogue 　 889
24.8 summary 　891
bibliographical and historical notes 　 892
exercises 　 894
25 machine translation　 895
25.1 why machine translation is hard 　 898
25.1.1 typology 　 898
25.1.2 other structural divergences 　　 900
25.1.3 lexicaldivergences 　 901
25.2 classical mt and the vauquois triangle 903
25.2.1 direct translation 　 904
25.2.2 transfer 　 906
25.2.3 combined direct and transfer approaches in classic mt　 908
25.2.4 the interlingua idea: using meaning 　　 909
25.3 statisticalmt 　 910
25.4 p(f e): the phrase-based translation model　　 913
25.5 alignment inmt 　 915
25.5.1 ibmmodel 1 　 916
25.5.2 hmmalignment 　　919
25.6 training alignment models　 921
25.6.1 em for training alignment models 　　922
25.7 symmetrizing alignments for phrase-based mt　 924
25.8 decoding for phrase-based statistical mt 　　 926
25.9 mtevaluation 　 930
25.9.1 using human raters 　 930
25.9.2 automatic evaluation: bleu 　　 931
25.10 advanced: syntactic models for mt 　　 934
25.11 advanced: ibm model 3 and fertility 　　935
25.11.1 training formodel 3 　939
25.12 advanced: log-linear models for mt 　　 939
25.13 summary 　940
bibliographical and historical notes 　 941
exercises 943
bibliography 　 945
author index　 995
subject index 　 1007