Morphological and Part-of-Speech Tagging of Historical Language - - PowerPoint PPT Presentation

▶
morphological and part of speech tagging of historical
SMART_READER_LITE
LIVE PREVIEW

Morphological and Part-of-Speech Tagging of Historical Language - - PowerPoint PPT Presentation

Dec 19, 2023 380 likes •995 views

Morphological and Part-of-Speech Tagging of Historical Language Data: A Comparison Stefanie Dipper Linguistics Department Ruhr-University Bochum 5 January, 2012 Workshop on Annotation of Corpora for Research in the Humanities Heidelberg

slide-1
SLIDE 1

Morphological and Part-of-Speech Tagging of Historical Language Data: A Comparison

Stefanie Dipper

Linguistics Department Ruhr-University Bochum

5 January, 2012 Workshop on Annotation of Corpora for Research in the Humanities Heidelberg

Stefanie Dipper Morphological and POS Tagging 5.1.2012 1 / 27

slide-2
SLIDE 2

Goals

A morphological and a part-of-speech (POS) tagger for texts from Middle High German (MHG, 1050–1350)

Ex: sach ‘saw’: 3.Sg.Past.Ind (Morph) VVFIN (POS)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 2 / 27

slide-3
SLIDE 3

Goals

A morphological and a part-of-speech (POS) tagger for texts from Middle High German (MHG, 1050–1350)

Ex: sach ‘saw’: 3.Sg.Past.Ind (Morph) VVFIN (POS)

Use of available state-of-the-art taggers

without any adaption some preprocessing of the annotated training data

Stefanie Dipper Morphological and POS Tagging 5.1.2012 2 / 27

slide-4
SLIDE 4

Goals

A morphological and a part-of-speech (POS) tagger for texts from Middle High German (MHG, 1050–1350)

Ex: sach ‘saw’: 3.Sg.Past.Ind (Morph) VVFIN (POS)

Use of available state-of-the-art taggers

without any adaption some preprocessing of the annotated training data

Challenge: highly variant data

no spelling conventions, e.g. sitzen, sizzen ‘sit’ different dialects, e.g. bruoder, pruder ‘brother’

→ Data sparseness

Stefanie Dipper Morphological and POS Tagging 5.1.2012 2 / 27

slide-5
SLIDE 5

Goals

A morphological and a part-of-speech (POS) tagger for texts from Middle High German (MHG, 1050–1350)

Ex: sach ‘saw’: 3.Sg.Past.Ind (Morph) VVFIN (POS)

Use of available state-of-the-art taggers

without any adaption some preprocessing of the annotated training data

Challenge: highly variant data

no spelling conventions, e.g. sitzen, sizzen ‘sit’ different dialects, e.g. bruoder, pruder ‘brother’

→ Data sparseness Questions:

(how much) does normalization help? – original (“diplomatic”) vs. normalized wordforms does POS preprocessing help morphological tagging?

Stefanie Dipper Morphological and POS Tagging 5.1.2012 2 / 27

slide-6
SLIDE 6

Outline

1

The corpus

2

Training experiments

Stefanie Dipper Morphological and POS Tagging 5.1.2012 3 / 27

slide-7
SLIDE 7

The corpus

Outline

1

The corpus

2

Training experiments

Stefanie Dipper Morphological and POS Tagging 5.1.2012 4 / 27

slide-8
SLIDE 8

The corpus

Project context

Projects “Mittelhochdeutsche Grammatik” and “Referenzkorpus Mittelhochdeutsch” (Universities of Bochum and Bonn)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 5 / 27

slide-9
SLIDE 9

The corpus

Project context

Projects “Mittelhochdeutsche Grammatik” and “Referenzkorpus Mittelhochdeutsch” (Universities of Bochum and Bonn) Goals

a balanced, annotated reference corpus of MHG diplomatic transcriptions final size: 300 texts, 2 million wordforms available via the internet (ANNIS)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 5 / 27

slide-10
SLIDE 10

The corpus

Project context

Projects “Mittelhochdeutsche Grammatik” and “Referenzkorpus Mittelhochdeutsch” (Universities of Bochum and Bonn) Goals

a balanced, annotated reference corpus of MHG diplomatic transcriptions final size: 300 texts, 2 million wordforms available via the internet (ANNIS)

Annotations

parts of speech (POS) morphological tags lemma normalized wordform

Stefanie Dipper Morphological and POS Tagging 5.1.2012 5 / 27

slide-11
SLIDE 11

The corpus

Project context

Projects “Mittelhochdeutsche Grammatik” and “Referenzkorpus Mittelhochdeutsch” (Universities of Bochum and Bonn) Goals

a balanced, annotated reference corpus of MHG diplomatic transcriptions final size: 300 texts, 2 million wordforms available via the internet (ANNIS)

Annotations

parts of speech (POS) morphological tags lemma normalized wordform

Currently: semi-automatic annotation

tools by Thomas Klein, Bonn (2001) require a lot of human intervention

Stefanie Dipper Morphological and POS Tagging 5.1.2012 5 / 27

slide-12
SLIDE 12

The corpus

The data

51 texts with 211,000 tokens from the MHG Reference Corpus From two dialect regions: Upper (UG) and Central German (CG)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 6 / 27

slide-13
SLIDE 13

The corpus

Upper and Central (and Lower) German

Source: Wikimedia

Stefanie Dipper Morphological and POS Tagging 5.1.2012 7 / 27

slide-14
SLIDE 14

The corpus

Spelling variation

E.g. (normalized) wolte ‘wanted’ – wolt – wolta – wolte – woltt – wolti – wolthe – walde – uuolde – volde – . . .

Stefanie Dipper Morphological and POS Tagging 5.1.2012 8 / 27

slide-15
SLIDE 15

The corpus

Spelling variation

E.g. (normalized) wolte ‘wanted’ – wolt – wolta – wolte – woltt – wolti – wolthe – walde – uuolde – volde – . . . Normalization: mapping to a virtual, idealized historical wordform

Stefanie Dipper Morphological and POS Tagging 5.1.2012 8 / 27

slide-16
SLIDE 16

The corpus

The data: some statistics

Texts Tokens Types diplomatic normalized [ NHG ] 51 total 211,000 40,500 20,500 .19 .10 [.14] 27 CG 91,000 22,000 13,000 .24 .14 [.18] 20 UG 67,000 15,000 8,500 .22 .13 4 mixed 53,000

Stefanie Dipper Morphological and POS Tagging 5.1.2012 9 / 27

slide-17
SLIDE 17

The corpus

The data: some statistics

Texts Tokens Types diplomatic normalized [ NHG ] 51 total 211,000 40,500 20,500 .19 .10 [.14] 27 CG 91,000 22,000 13,000 .24 .14 [.18] 20 UG 67,000 15,000 8,500 .22 .13 4 mixed 53,000

On average: roughly 2 spelling variants (diplomatic) per wordform (normalized)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 9 / 27

slide-18
SLIDE 18

The corpus

The data: some statistics

Texts Tokens Types diplomatic normalized [ NHG ] 51 total 211,000 40,500 20,500 .19 .10 [.14] 27 CG 91,000 22,000 13,000 .24 .14 [.18] 20 UG 67,000 15,000 8,500 .22 .13 4 mixed 53,000

On average: roughly 2 spelling variants (diplomatic) per wordform (normalized) Type-token ratio: higher ratio → more diverse data – CG more diverse than UG [– cf. modern German (NHG): TTR = .14/.18]

Stefanie Dipper Morphological and POS Tagging 5.1.2012 9 / 27

slide-19
SLIDE 19

The corpus

Predictions I

  • 1. Normalized vs. diplomatic:

tagging normalized data should be easier

Stefanie Dipper Morphological and POS Tagging 5.1.2012 10 / 27

slide-20
SLIDE 20

The corpus

Predictions I

  • 1. Normalized vs. diplomatic:

tagging normalized data should be easier

  • 2. CG vs. UG vs. NHG: rather unclear

a) pro CG: more training data available b) pro UG: less diverse (lower type/token ratio) c) pro MHG (normalized, equal size): less diverse than NHG

Stefanie Dipper Morphological and POS Tagging 5.1.2012 10 / 27

slide-21
SLIDE 21

The corpus

Tagsets

POS based on the STTS tagset (standard German tagset) NN, NE “normal noun, proper noun” VVFIN, VVINF, VVIMP, VVPP “finite full verb, infinitive, imperative, past participle”

Stefanie Dipper Morphological and POS Tagging 5.1.2012 11 / 27

slide-22
SLIDE 22

The corpus

Tagsets

POS based on the STTS tagset (standard German tagset) NN, NE “normal noun, proper noun” VVFIN, VVINF, VVIMP, VVPP “finite full verb, infinitive, imperative, past participle” Morphology “large” STTS tagset Comp.Fem.Acc.Sg “(adjective:) comparative form, feminine, accusative, singular” 3.Sg.Past.* “(verb:) 3rd singular past tense, unspecified for mood”

Stefanie Dipper Morphological and POS Tagging 5.1.2012 11 / 27

slide-23
SLIDE 23

The corpus

Underspecification

POS and morph: more underspecified tags in MHG than in NHG (no native speakers) Gender of nouns: not yet as fixed as nowadays

Example: slange ‘snake’: masc/fem daz si slangen bizzen

  • *.Acc.Pl

MascFem.Nom.Pl 3.Pl.Past.* that them snakes bit ‘that snakes bit them’

Stefanie Dipper Morphological and POS Tagging 5.1.2012 12 / 27

slide-24
SLIDE 24

The corpus

Tagsets: some statistics (normalized data)

POS # Tags Ø Tags/wofo Median (max) CG norm 44 1.10 ± 0.37 1 (7) UG norm 41 1.10 ± 0.35 1 (6) NHG (210K) 53 1.05 ± 0.23 1 (6) (90K) 51 1.04 ± 0.21 1 (6)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 13 / 27

slide-25
SLIDE 25

The corpus

Tagsets: some statistics (normalized data)

POS # Tags Ø Tags/wofo Median (max) CG norm 44 1.10 ± 0.37 1 (7) UG norm 41 1.10 ± 0.35 1 (6) NHG (210K) 53 1.05 ± 0.23 1 (6) (90K) 51 1.04 ± 0.21 1 (6) Morphology # Tags Ø Tags/wofo Median (max) CG norm 245 1.40 ± 1.16 1 (23) UG norm 219 1.46 ± 1.28 1 (33) NHG (210K) 230 1.37 ± 0.97 1 (26) (90K) 205 1.32 ± 0.86 1 (18)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 13 / 27

slide-26
SLIDE 26

The corpus

Predictions II

  • 1. Normalized vs. diplomatic:

tagging normalized data should be easier

  • 2. CG vs. UG vs. NHG: rather unclear

a) pro CG: more training data available b) pro UG: less diverse (lower type/token ratio) c) pro MHG (normalized, equal size): less diverse than NHG

  • 3. Morphology vs. POS:

tagging POS should be easier (lower ambiguity rate)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 14 / 27

slide-27
SLIDE 27

The corpus

Predictions II

  • 1. Normalized vs. diplomatic:

tagging normalized data should be easier

  • 2. CG vs. UG vs. NHG: rather unclear

a) pro CG: more training data available b) pro UG: less diverse (lower type/token ratio) c) pro MHG (normalized, equal size): less diverse than NHG

  • 3. Morphology vs. POS:

tagging POS should be easier (lower ambiguity rate)

  • 4. CG vs. UG vs. NHG: again, rather unclear

a) CG and UG rather similar b) POS: pro UG, morph: pro CG (lower maxima) c) pro NHG: lower ambiguity rates

Stefanie Dipper Morphological and POS Tagging 5.1.2012 14 / 27

slide-28
SLIDE 28

Training experiments

Outline

1

The corpus

2

Training experiments

Stefanie Dipper Morphological and POS Tagging 5.1.2012 15 / 27

slide-29
SLIDE 29

Training experiments

Other approaches

Usually: map the historical wordforms to modern wordforms and apply a modern (trained) tagger (e.g. Rayson et al. 2007, Pilz et

  • al. 2006)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 16 / 27

slide-30
SLIDE 30

Training experiments

Other approaches

Usually: map the historical wordforms to modern wordforms and apply a modern (trained) tagger (e.g. Rayson et al. 2007, Pilz et

  • al. 2006)

Here: train and apply a tagger to historical wordforms (diplomatic and normalized)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 16 / 27

slide-31
SLIDE 31

Training experiments

The tagger: TreeTagger (Schmid 1994, 1995)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 17 / 27

slide-32
SLIDE 32

Training experiments

The tagger: TreeTagger (Schmid 1994, 1995)

Training on annotated data

Stefanie Dipper Morphological and POS Tagging 5.1.2012 17 / 27

slide-33
SLIDE 33

Training experiments

The tagger: TreeTagger (Schmid 1994, 1995)

Training on annotated data TreeTagger assigns tags on the basis of

1

a fullform lexicon e.g. saw: VERB see NOUN saw

2

prefix and suffix information e.g. -ous → 96% ADJ, 4% NOUN

3

context information from n preceding tags (bigrams, trigrams, . . . ) e.g. DET ADJ ? → 70% NOUN, 10% ADJ, . . .

Stefanie Dipper Morphological and POS Tagging 5.1.2012 17 / 27

slide-34
SLIDE 34

Training experiments

The tagger: TreeTagger (Schmid 1994, 1995)

Training on annotated data TreeTagger assigns tags on the basis of

1

a fullform lexicon e.g. saw: VERB see NOUN saw

2

prefix and suffix information e.g. -ous → 96% ADJ, 4% NOUN

3

context information from n preceding tags (bigrams, trigrams, . . . ) e.g. DET ADJ ? → 70% NOUN, 10% ADJ, . . .

Performance:

97.53% accuracy for modern German (POS tagging) best published tagging accuracy for German (as of 2009)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 17 / 27

slide-35
SLIDE 35

Training experiments

The tagger: TreeTagger (Schmid 1994, 1995)

Training on annotated data TreeTagger assigns tags on the basis of

1

a fullform lexicon e.g. saw: VERB see NOUN saw

2

prefix and suffix information e.g. -ous → 96% ADJ, 4% NOUN

3

context information from n preceding tags (bigrams, trigrams, . . . ) e.g. DET ADJ ? → 70% NOUN, 10% ADJ, . . .

Performance:

97.53% accuracy for modern German (POS tagging) best published tagging accuracy for German (as of 2009)

Easy to get and to use

Stefanie Dipper Morphological and POS Tagging 5.1.2012 17 / 27

slide-36
SLIDE 36

Training experiments

POS and morphology experiments: 3 parameters

Stefanie Dipper Morphological and POS Tagging 5.1.2012 18 / 27

slide-37
SLIDE 37

Training experiments

POS and morphology experiments: 3 parameters

1

CG vs. UG data

Stefanie Dipper Morphological and POS Tagging 5.1.2012 18 / 27

slide-38
SLIDE 38

Training experiments

POS and morphology experiments: 3 parameters

1

CG vs. UG data

2

Diplomatic vs. normalized wordforms

Stefanie Dipper Morphological and POS Tagging 5.1.2012 18 / 27

slide-39
SLIDE 39

Training experiments

POS and morphology experiments: 3 parameters

1

CG vs. UG data

2

Diplomatic vs. normalized wordforms

3

Training on all data vs. dialect-specific training (“generic” vs. “specific” taggers)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 18 / 27

slide-40
SLIDE 40

Training experiments

Results (summary)

Top results

POS: 92.91% (UG norm, specific) Morph: 80.84% (CG norm, generic)

  • cf. NHG (210K): 95.67% (POS), 76.95% (morph)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 19 / 27

slide-41
SLIDE 41

Training experiments

Results (summary)

Top results

POS: 92.91% (UG norm, specific) Morph: 80.84% (CG norm, generic)

  • cf. NHG (210K): 95.67% (POS), 76.95% (morph)

→ Predictions 2/4: results mirror ambiguity rates → Prediction 3 is confirmed: gap of > 10 percentage points between POS and morph

Stefanie Dipper Morphological and POS Tagging 5.1.2012 19 / 27

slide-42
SLIDE 42

Training experiments

Results (summary)

Top results

POS: 92.91% (UG norm, specific) Morph: 80.84% (CG norm, generic)

  • cf. NHG (210K): 95.67% (POS), 76.95% (morph)

→ Predictions 2/4: results mirror ambiguity rates → Prediction 3 is confirmed: gap of > 10 percentage points between POS and morph

Normalization: considerable improvements

4–7 percentage points (p < .001) → Prediction 1 is confirmed

Stefanie Dipper Morphological and POS Tagging 5.1.2012 19 / 27

slide-43
SLIDE 43

Training experiments

Results (summary)

Top results

POS: 92.91% (UG norm, specific) Morph: 80.84% (CG norm, generic)

  • cf. NHG (210K): 95.67% (POS), 76.95% (morph)

→ Predictions 2/4: results mirror ambiguity rates → Prediction 3 is confirmed: gap of > 10 percentage points between POS and morph

Normalization: considerable improvements

4–7 percentage points (p < .001) → Prediction 1 is confirmed

Generic vs. specific taggers: more (but heterogeneous) training data helps (significant in most scenarios)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 19 / 27

slide-44
SLIDE 44

Training experiments

Morphology experiment: an additional parameter

Integrate POS information:

Stefanie Dipper Morphological and POS Tagging 5.1.2012 20 / 27

slide-45
SLIDE 45

Training experiments

Morphology experiment: an additional parameter

Integrate POS information:

1

Original version: no use of POS; bigrams werde 3.Sg.Pres.Subj disemo Neut.Dat.Sg ‘would this’

Stefanie Dipper Morphological and POS Tagging 5.1.2012 20 / 27

slide-46
SLIDE 46

Training experiments

Morphology experiment: an additional parameter

Integrate POS information:

1

Original version: no use of POS; bigrams werde 3.Sg.Pres.Subj disemo Neut.Dat.Sg ‘would this’

2

Successive pairs of <wofo, POS><wofo, morph>; trigrams werde VAFIN werde 3.Sg.Pres.Subj disemo PD disemo Neut.Dat.Sg

Stefanie Dipper Morphological and POS Tagging 5.1.2012 20 / 27

slide-47
SLIDE 47

Training experiments

Morphology experiment: an additional parameter

Integrate POS information:

1

Original version: no use of POS; bigrams werde 3.Sg.Pres.Subj disemo Neut.Dat.Sg ‘would this’

2

Successive pairs of <wofo, POS><wofo, morph>; trigrams werde VAFIN werde 3.Sg.Pres.Subj disemo PD disemo Neut.Dat.Sg

3

Merged pairs of <wofo.POS, morph>; bigrams werde.VAFIN 3.Sg.Pres.Subj disemo.PD Neut.Dat.Sg

Stefanie Dipper Morphological and POS Tagging 5.1.2012 20 / 27

slide-48
SLIDE 48

Training experiments

Morphology experiment: an additional parameter

Integrate POS information:

1

Original version: no use of POS; bigrams werde 3.Sg.Pres.Subj disemo Neut.Dat.Sg ‘would this’

2

Successive pairs of <wofo, POS><wofo, morph>; trigrams werde VAFIN werde 3.Sg.Pres.Subj disemo PD disemo Neut.Dat.Sg

3

Merged pairs of <wofo.POS, morph>; bigrams werde.VAFIN 3.Sg.Pres.Subj disemo.PD Neut.Dat.Sg

4

(Upper bound: merged pairs with gold POS)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 20 / 27

slide-49
SLIDE 49

Training experiments

Sample tagging rules

Successive pairs: tag[-1] = ART tag[-2] = Masc.Nom.Sg Masc.Nom.Sg 0.625717 Pos.Masc.Nom.Sg 0.207722

Stefanie Dipper Morphological and POS Tagging 5.1.2012 21 / 27

slide-50
SLIDE 50

Training experiments

Sample tagging rules

Successive pairs: tag[-1] = ART tag[-2] = Masc.Nom.Sg Masc.Nom.Sg 0.625717 Pos.Masc.Nom.Sg 0.207722 Merged pairs: n.PIS *.Nom.Sg 0.863558 *.Acc.Sg 0.106894 *.*.* 0.029548 e.PIS *.Dat.Sg 0.831818 *.Nom.Sg 0.090909 *.Acc.Sg 0.077273

Stefanie Dipper Morphological and POS Tagging 5.1.2012 21 / 27

slide-51
SLIDE 51

Training experiments

Exemplary results

Results with CG norm, generic: (i) No use 79.70 (ii) Successive pairs 80.84 (iii) Merged pairs 79.81 (iv) Gold POS 82.19 Some improvements with successive pairs No improvement with merged pairs

Stefanie Dipper Morphological and POS Tagging 5.1.2012 22 / 27

slide-52
SLIDE 52

Training experiments

Summary and Outlook

Stefanie Dipper Morphological and POS Tagging 5.1.2012 23 / 27

slide-53
SLIDE 53

Training experiments

Summary and Outlook

POS tagging

satisfiable results (> 91%) clearly better results with modern data (modern data: no cross-validation)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 23 / 27

slide-54
SLIDE 54

Training experiments

Summary and Outlook

POS tagging

satisfiable results (> 91%) clearly better results with modern data (modern data: no cross-validation)

Morphological tagging

needs more sophisticated tagging methods (> 79%) e.g. RFTagger (Schmid & Laws 2008): analyzes complex morphological tags better results with historical data (maybe due to greater extent of underspecified annotations)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 23 / 27

slide-55
SLIDE 55

Training experiments

Summary and Outlook

POS tagging

satisfiable results (> 91%) clearly better results with modern data (modern data: no cross-validation)

Morphological tagging

needs more sophisticated tagging methods (> 79%) e.g. RFTagger (Schmid & Laws 2008): analyzes complex morphological tags better results with historical data (maybe due to greater extent of underspecified annotations)

Normalization: increases accuracy by 4–7 percentage points (with POS and morphological tagging)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 23 / 27

slide-56
SLIDE 56

Training experiments

Summary and Outlook

POS tagging

satisfiable results (> 91%) clearly better results with modern data (modern data: no cross-validation)

Morphological tagging

needs more sophisticated tagging methods (> 79%) e.g. RFTagger (Schmid & Laws 2008): analyzes complex morphological tags better results with historical data (maybe due to greater extent of underspecified annotations)

Normalization: increases accuracy by 4–7 percentage points (with POS and morphological tagging) TODO: Error analysis

Stefanie Dipper Morphological and POS Tagging 5.1.2012 23 / 27

slide-57
SLIDE 57

Training experiments

Thank you!

Stefanie Dipper Morphological and POS Tagging 5.1.2012 24 / 27

slide-58
SLIDE 58

Training experiments

(Semi-)automatic annotation

How many correct tags are among the top n most probable tags? Part of Speech Rank # Word forms 1 8160 92.0% 2 370 4.2% 3 27 0.3% None 303 3.4% Morphology (merged) Rank # Word forms 1 7467 79.6% 2 600 6.4% 3 99 1.1% None 1122 12.0% → Top-3 ranks: 96.5% (POS), 87.1% (morph)

Stefanie Dipper Morphological and POS Tagging 5.1.2012 25 / 27

slide-59
SLIDE 59

Training experiments

POS experiments: results

Dialect Tagger Word Forms diplomatic normalized [NHG] CG generic 86.92 91.66 [95.67] specific 86.62 91.43 [94.39] UG generic 88.88 92.83 specific 89.16 92.91

Stefanie Dipper Morphological and POS Tagging 5.1.2012 26 / 27

slide-60
SLIDE 60

Training experiments

Morphological experiments: results

Scenario Dialect Tagger Word Forms diplomatic normalized [NHG] (i) No use CG gen 73.91 79.70 [76.95] spec 72.64 78.43 [75.71] UG gen 73.85 78.28 spec 73.23 78.15 (ii) Succ. pairs CG gen 74.23 80.84 spec 72.37 79.47 UG gen 74.17 79.11 spec 73.27 78.63 (iii) Merged pairs CG gen 74.39 79.81 spec 72.86 78.48 UG gen 74.07 77.63 spec 73.14 77.02 (iv) Gold POS CG gen 77.14 82.19 spec 75.54 80.80 UG gen 76.79 80.83 spec 75.79 80.26

Stefanie Dipper Morphological and POS Tagging 5.1.2012 27 / 27