Rhetorical structure and argumentation structure in monologue text - - PowerPoint PPT Presentation

▶
rhetorical structure and argumentation structure in
SMART_READER_LITE
LIVE PREVIEW

Rhetorical structure and argumentation structure in monologue text - - PowerPoint PPT Presentation

Jan 15, 2023 41 likes •614 views

Rhetorical structure and argumentation structure in monologue text Andreas Peldszus Manfred Stede Applied Computational Linguistics, University of Potsdam 3rd Workshop on Argument Mining @ACL 2016, Berlin, 12.08.2016 Peldszus, Stede (Uni

slide-1
SLIDE 1

Rhetorical structure and argumentation structure in monologue text

Andreas Peldszus Manfred Stede

Applied Computational Linguistics, University of Potsdam

3rd Workshop on Argument Mining @ACL 2016, Berlin, 12.08.2016

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 1 / 24

slide-2
SLIDE 2

Outline

1 Introduction 2 Matching RST and argumentation: Qualitative analysis 3 Automatically deriving ARG from RST

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 2 / 24

slide-3
SLIDE 3

Outline

1 Introduction 2 Matching RST and argumentation: Qualitative analysis 3 Automatically deriving ARG from RST

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 3 / 24

slide-4
SLIDE 4

RST in a nutshell

Key ideas and principles [Mann and Thompson, 1988]

  • text coherent <=> a plausible RST tree exists
  • 25 relations: presentational (pragmatic) vs.

subject-matter (semantic)

  • most relations: nucleus (main info/act) + satellite

(support info/act)

  • same relation set applies to minimal units and

recursively to text spans

  • every unit/span takes part in the analysis
  • no crossing edges
  • (annotation guidelines in [Stede, 2016])

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 4 / 24

slide-5
SLIDE 5

Argumentation structure in a nutshell

[e1] Health insurance companies should naturally cover alternative medical treatments. [e2] Not all practices and approaches that are lumped together under this term may have been proven in clinical trials,

1

[e3] yet it's precisely their positive effect when accompanying conventional 'western' medical therapies that's been demonstrated as beneficial.

2

[e4] Besides many general practitioners offer such counselling and treatments in parallel anyway -

3

[e5] and who would want to question their broad expertise?

4 5

c3 c4 c2

Freeman’s theory, revised & slightly generalized:

[Freeman, 1991, 2011] [Peldszus and Stede, 2013]

  • node types = argumentative role

proponent (presents and defends claims)

  • pponent (critically questions)
  • link types = argumentative function

support own claims (normally, by example) attack other’s claims (rebut, undercut)

  • (annotation guidelines in [Stede, 2016])

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 5 / 24

slide-6
SLIDE 6

Outline

1 Introduction 2 Matching RST and argumentation: Qualitative analysis 3 Automatically deriving ARG from RST

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 6 / 24

slide-7
SLIDE 7

Dataset: argumentative microtexts

Properties:

  • about 5 segments long
  • each segment is arg. relevant
  • explicit main claim
  • at least one possible objection considered

Texts:

  • 23 texts: hand-crafted, covering different arg. configurations
  • 92 texts: collected in a controlled text generation experiment
  • with professional parallel translation to English
  • all annotated with argumentation structure
  • freely available, CC-by-nc-sa license; see [Peldszus and Stede, 2016]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 7 / 24

slide-8
SLIDE 8

Dataset: argumentative microtexts

Properties:

  • about 5 segments long
  • each segment is arg. relevant
  • explicit main claim
  • at least one possible objection considered

Texts:

  • 23 texts: hand-crafted, covering different arg. configurations
  • 92 texts: collected in a controlled text generation experiment
  • with professional parallel translation to English
  • all annotated with argumentation structure
  • freely available, CC-by-nc-sa license; see [Peldszus and Stede, 2016]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 7 / 24

slide-9
SLIDE 9

Multi-layer discourse annotation

How does argumentation structure relate to other discourse structures?

  • Rhetorical Structure Theory (RST)

[Mann and Thompson, 1988]

  • Segmented Discourse Structure Theory (SDRT)

[Asher and Lascarides, 2003]

Joint work with Stergos Afantenos, Nicholas Asher, Jérémy Perret

[Stede et al., 2016]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 8 / 24

slide-10
SLIDE 10

Multi-layer discourse annotation: Harmonize segmentation

[e1] Supermarket employees and people who work in shopping centres also have the right to a Sunday off work. [e2] Likewise public holidays should remain what they are:

1

[e3] for some a day

  • f introspection,

for others a paid day off that is not taken away from the annual paid leave proper.

2+3

[e4] Hence it is good when shops are not open on Sundays and public holidays. [e5] People, however, who work during the week and

  • n Saturdays then

have a problem:

4

[e6] everyone else can shop weekdays,

5+6+7

[e7] but they can't. [e8] For those people the late

  • pening hours, which

meanwhile already extend to 12:00 midnight, present a good alternative.

8 2+3 5+6+7

c12 c11 c14 c13

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 9 / 24

slide-11
SLIDE 11

Qualitative: Central Claim

Total: 115 CCs in ARG (one per text)

  • Canonical: In 95 texts (85%), central nucleus in RST corresponds to central claim in ARG
  • In 5 texts, they are disjoint
  • multiple statements of the CC
  • no explicit CC
  • In 12 texts, they overlap
  • ARG CC has more fine-grained RST analysis (e.g., Condition)
  • multinuclear RST relations yield multiple RSTnuc for the text

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 10 / 24

slide-12
SLIDE 12

Qualitative: Support

Total: 261 Support relations in ARG

  • Canonical: 132 correspond to RST Reason, Justify, Evidence, Motivation, Cause
  • But: 77% of the texts contain at least one non-canonical Support
  • 12 Supports correspond to another (mostly ‘informational’) RST relation
  • 117 Supports have no corresponding RST relation
  • RST segment is in a multinuclear relation (70)
  • RST segment is related to a different segment via an informational relation (21)
  • Mismatch in Support transitivity (16)
  • Other (18)

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 11 / 24

slide-13
SLIDE 13

Qualitative: Attack

Total: 98 Attack relations in ARG

  • Simple: A single attacking node (either leaf or supported)
  • Canonical: (24/31) Attack corresponds to Antithesis, Contrast, Concession
  • (7/31) opponent voice absent in RST, or segment connected otherwise
  • Medium: Multiple individual attacks in ARG
  • Canonical: In all 7 cases, RST groups them via Conjunction
  • Complex: Attack and Counterattack
  • Canonical: (47/60) Attack corresponds to a backward Concession, Antithesis

(different levels of complexity)

  • (13/60) Annotator did not see this argumentative function as primary

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 12 / 24

slide-14
SLIDE 14

Qualitative: Attack

Total: 98 Attack relations in ARG

  • Simple: A single attacking node (either leaf or supported)
  • Canonical: (24/31) Attack corresponds to Antithesis, Contrast, Concession
  • (7/31) opponent voice absent in RST, or segment connected otherwise
  • Medium: Multiple individual attacks in ARG
  • Canonical: In all 7 cases, RST groups them via Conjunction
  • Complex: Attack and Counterattack
  • Canonical: (47/60) Attack corresponds to a backward Concession, Antithesis

(different levels of complexity)

  • (13/60) Annotator did not see this argumentative function as primary

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 12 / 24

slide-15
SLIDE 15

Qualitative: Attack

Total: 98 Attack relations in ARG

  • Simple: A single attacking node (either leaf or supported)
  • Canonical: (24/31) Attack corresponds to Antithesis, Contrast, Concession
  • (7/31) opponent voice absent in RST, or segment connected otherwise
  • Medium: Multiple individual attacks in ARG
  • Canonical: In all 7 cases, RST groups them via Conjunction
  • Complex: Attack and Counterattack
  • Canonical: (47/60) Attack corresponds to a backward Concession, Antithesis

(different levels of complexity)

  • (13/60) Annotator did not see this argumentative function as primary

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 12 / 24

slide-16
SLIDE 16

Outline

1 Introduction 2 Matching RST and argumentation: Qualitative analysis 3 Automatically deriving ARG from RST

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 13 / 24

slide-17
SLIDE 17

Task

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 14 / 24

slide-18
SLIDE 18

Task

⇒

[e1] Health insurance companies should naturally cover alternative medical treatments. [e2] Not all practices and approaches that are lumped together under this term may have been proven in clinical trials,

1

[e3] yet it's precisely their positive effect when accompanying conventional 'western' medical therapies that's been demonstrated as beneficial.

2

[e4] Besides many general practitioners

  • ffer such

counselling and treatments in parallel anyway

  • 3

[e5] and who would want to question their broad expertise?

4 5

c9 c7 c6

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 14 / 24

slide-19
SLIDE 19

Task

1 2 3 4 5

concession reason reason joint

⇒

1 2 3 4 5

rebut undercut support link

Common dependency format [Stede et al., 2016]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 14 / 24

slide-20
SLIDE 20

Evaluation procedure

Evaluate four aspects of the predicted structure:

  • central claim (cc): [yes, no]
  • role (ro): [proponent, opponent]
  • function (fu): [support, example, rebut,

undercut, link, join]

  • attachment (at): [yes, no]

1 2 3 4 5

rebut undercut support link

Procedure and train/test splits as in [Peldszus and Stede, 2015]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 15 / 24

slide-21
SLIDE 21

Evaluation procedure

Evaluate four aspects of the predicted structure:

  • central claim (cc): [yes, no]
  • role (ro): [proponent, opponent]
  • function (fu): [support, example, rebut,

undercut, link, join]

  • attachment (at): [yes, no]

1 2 3 4 5

rebut undercut support link root

Procedure and train/test splits as in [Peldszus and Stede, 2015]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 15 / 24

slide-22
SLIDE 22

Evaluation procedure

Evaluate four aspects of the predicted structure:

  • central claim (cc): [yes, no]
  • role (ro): [proponent, opponent]
  • function (fu): [support, example, rebut,

undercut, link, join]

  • attachment (at): [yes, no]

1 2 3 4 5

rebut undercut support link

Procedure and train/test splits as in [Peldszus and Stede, 2015]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 15 / 24

slide-23
SLIDE 23

Evaluation procedure

Evaluate four aspects of the predicted structure:

  • central claim (cc): [yes, no]
  • role (ro): [proponent, opponent]
  • function (fu): [support, example, rebut,

undercut, link, join]

  • attachment (at): [yes, no]

1 2 3 4 5

rebut undercut support link

Procedure and train/test splits as in [Peldszus and Stede, 2015]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 15 / 24

slide-24
SLIDE 24

Evaluation procedure

Evaluate four aspects of the predicted structure:

  • central claim (cc): [yes, no]
  • role (ro): [proponent, opponent]
  • function (fu): [support, example, rebut,

undercut, link, join]

  • attachment (at): [yes, no]

1 2 3 4 5

rebut undercut support link

Procedure and train/test splits as in [Peldszus and Stede, 2015]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 15 / 24

slide-25
SLIDE 25

Model 1: Heuristic baseline (BL)

Procedure:

1 predict ARG structure isomorphic to RST tree 2 map RST relations to ARG relation

best correlation according to [Stede et al., 2016]

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 16 / 24

slide-26
SLIDE 26

Model 1: Heuristic baseline (BL)

Procedure:

1 predict ARG structure isomorphic to RST tree 2 map RST relations to ARG relation

best correlation according to [Stede et al., 2016]

1 2 3 4 5

concession reason reason joint

⇓

1 2 3 4 5

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 16 / 24

slide-27
SLIDE 27

Model 1: Heuristic baseline (BL)

Procedure:

1 predict ARG structure isomorphic to RST tree 2 map RST relations to ARG relation

best correlation according to [Stede et al., 2016]

support: background, cause, evidence, justify, list, motiva- tion, reason, restatement, result rebut: antithesis, contrast, unless undercut: concession join: circumstance, condition, conjunction, disjunction, e- elaboration, elaboration, evaluation-s, evaluation-n, interpre- tation*, joint, means, preparation, purpose, sameunit, solu- tionhood* 1 2 3 4 5

concession reason reason joint

⇓

1 2 3 4 5

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 16 / 24

slide-28
SLIDE 28

Model 1: Heuristic baseline (BL)

Procedure:

1 predict ARG structure isomorphic to RST tree 2 map RST relations to ARG relation

best correlation according to [Stede et al., 2016]

support: background, cause, evidence, justify, list, motiva- tion, reason, restatement, result rebut: antithesis, contrast, unless undercut: concession join: circumstance, condition, conjunction, disjunction, e- elaboration, elaboration, evaluation-s, evaluation-n, interpre- tation*, joint, means, preparation, purpose, sameunit, solu- tionhood* 1 2 3 4 5

concession reason reason joint

⇓

1 2 3 4 5

undercut support support join

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 16 / 24

slide-29
SLIDE 29

Model 2: Naive aligner (A)

Training procedure:

1 find common connected

components

2 extract corresponding subgraphs 3 measure predictive probability

  • 1

2 3 4 5

concession reason reason joint

,

1 2 3 4 5

rebut undercut support link

  • Peldszus, Stede (Uni Potsdam)

Rhetorical structure and argumentation structure ArgMin WS 3 17 / 24

slide-30
SLIDE 30

Model 2: Naive aligner (A)

Training procedure:

1 find common connected

components

2 extract corresponding subgraphs 3 measure predictive probability

  • 1

2 3 4 5

concession reason reason joint

,

1 2 3 4 5

rebut undercut support link

  • Peldszus, Stede (Uni Potsdam)

Rhetorical structure and argumentation structure ArgMin WS 3 17 / 24

slide-31
SLIDE 31

Model 2: Naive aligner (A)

Training procedure:

1 find common connected

components

2 extract corresponding subgraphs 3 measure predictive probability

  • 1

2 3 4 5

concession reason reason joint

,

1 2 3 4 5

rebut undercut support link

  • a

b

joint

= ⇒

a b

link

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 17 / 24

slide-32
SLIDE 32

Model 2: Naive aligner (A)

Training procedure:

1 find common connected

components

2 extract corresponding subgraphs 3 measure predictive probability

  • 1

2 3 4 5

concession reason reason joint

,

1 2 3 4 5

rebut undercut support link

  • a

b

joint

p=0.7

= = = ⇒

a b

link

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 17 / 24

slide-33
SLIDE 33

Model 2: Naive aligner (A)

Training procedure:

1 find common connected

components

2 extract corresponding subgraphs 3 measure predictive probability

  • 1

2 3 4 5

concession reason reason joint

,

1 2 3 4 5

rebut undercut support link

  • a

b

joint

p=0.7

= = = ⇒

a b

link

a b c

reason joint

p=0.6

= = = ⇒

a b c

support link

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 17 / 24

slide-34
SLIDE 34

Model 2: Naive aligner (A)

Training procedure:

1 find common connected

components

2 extract corresponding subgraphs 3 measure predictive probability

  • 1

2 3 4 5

concession reason reason joint

,

1 2 3 4 5

rebut undercut support link

  • a

b

joint

p=0.7

= = = ⇒

a b

link

a b c

reason joint

p=0.6

= = = ⇒

a b c

support link

a b c

concession reason

p=0.5

= = = ⇒

a b c

rebut undercut

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 17 / 24

slide-35
SLIDE 35

Model 2: Naive aligner (A)

Testing procedure:

1 extract all subgraphs 2 look them up in the model 3 accumulate edge probabilities 4 decode with Minimum

Spanning Tree algorithm

[Chu and Liu, 1965, Edmonds, 1967] 1 2 3 4 5

concession reason reason joint

⇓

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 18 / 24

slide-36
SLIDE 36

Model 2: Naive aligner (A)

Testing procedure:

1 extract all subgraphs 2 look them up in the model 3 accumulate edge probabilities 4 decode with Minimum

Spanning Tree algorithm

[Chu and Liu, 1965, Edmonds, 1967] 1 2 3 4 5

concession reason reason joint

⇓

a b

concession

a b

reason

a b

joint

a b c

concession reason

a b c

reason reason

. . .

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 18 / 24

slide-37
SLIDE 37

Model 2: Naive aligner (A)

Testing procedure:

1 extract all subgraphs 2 look them up in the model 3 accumulate edge probabilities 4 decode with Minimum

Spanning Tree algorithm

[Chu and Liu, 1965, Edmonds, 1967] 1 2 3 4 5

concession reason reason joint

⇓

a b

undercut

a b

support

a b

link

a b c

rebut undercut

a b c

support support

. . .

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 18 / 24

slide-38
SLIDE 38

Model 2: Naive aligner (A)

Testing procedure:

1 extract all subgraphs 2 look them up in the model 3 accumulate edge probabilities 4 decode with Minimum

Spanning Tree algorithm

[Chu and Liu, 1965, Edmonds, 1967] 1 2 3 4 5

concession reason reason joint

⇓

1 2 3 4 5

rebut undercut support support link

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 18 / 24

slide-39
SLIDE 39

Model 2: Naive aligner (A)

Testing procedure:

1 extract all subgraphs 2 look them up in the model 3 accumulate edge probabilities 4 decode with Minimum

Spanning Tree algorithm

[Chu and Liu, 1965, Edmonds, 1967] 1 2 3 4 5

concession reason reason joint

⇓

1 2 3 4 5

support rebut undercut support link

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 18 / 24

slide-40
SLIDE 40

Model 2: Naive aligner (A)

Testing procedure:

1 extract all subgraphs 2 look them up in the model 3 accumulate edge probabilities 4 decode with Minimum

Spanning Tree algorithm

[Chu and Liu, 1965, Edmonds, 1967]

Note:

  • unconnected predictions:

initialize graph with low scored default edges

  • variant: enforce root of the

RST tree

1 2 3 4 5

concession reason reason joint

⇓

1 2 3 4 5

rebut undercut link

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 18 / 24

slide-41
SLIDE 41

Model 2: Naive aligner (A)

Testing procedure:

1 extract all subgraphs 2 look them up in the model 3 accumulate edge probabilities 4 decode with Minimum

Spanning Tree algorithm

[Chu and Liu, 1965, Edmonds, 1967]

Note:

  • unconnected predictions:

initialize graph with low scored default edges

  • variant: enforce root of the

RST tree

1 2 3 4 5

concession reason reason joint

⇓

1 2 3 4 5

rebut undercut support link root

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 18 / 24

slide-42
SLIDE 42

Model 3: Evidence graph (EG)

Evidence graph model [Peldszus and Stede, 2015]:

  • train one base classifier for each of the 4

levels (cc, ro, fu, at)

  • jointly predict all levels by combining the

predictions into one edge score

  • decode with MST

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 19 / 24

slide-43
SLIDE 43

Model 3: Evidence graph (EG)

Evidence graph model [Peldszus and Stede, 2015]:

  • train one base classifier for each of the 4

levels (cc, ro, fu, at)

  • jointly predict all levels by combining the

predictions into one edge score

  • decode with MST

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 19 / 24

slide-44
SLIDE 44

Model 3: Evidence graph (EG)

Evidence graph model [Peldszus and Stede, 2015]:

  • train one base classifier for each of the 4

levels (cc, ro, fu, at)

  • jointly predict all levels by combining the

predictions into one edge score

  • decode with MST

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 19 / 24

slide-45
SLIDE 45

Model 3: Evidence graph (EG)

Evidence graph model [Peldszus and Stede, 2015]:

  • train one base classifier for each of the 4

levels (cc, ro, fu, at)

  • jointly predict all levels by combining the

predictions into one edge score

  • decode with MST

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 19 / 24

slide-46
SLIDE 46

Model 3: Evidence graph (EG)

Segment feature sets:

  • base features incl. 2-node subgraph features:
  • position of the segment in the text
  • is it the first or the last segment?
  • has it incoming/outgoing edges?
  • number of incoming/outgoing edges
  • type of incoming/outgoing edges
  • 3-node subgraph features
  • all relation chains of length 2 involving this segment
  • 4-node subgraph features
  • all relation chains of length 3 involving this segment

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 20 / 24

slide-47
SLIDE 47

Model 3: Evidence graph (EG)

Segment feature sets:

  • base features incl. 2-node subgraph features:
  • position of the segment in the text
  • is it the first or the last segment?
  • has it incoming/outgoing edges?
  • number of incoming/outgoing edges
  • type of incoming/outgoing edges
  • 3-node subgraph features
  • all relation chains of length 2 involving this segment
  • 4-node subgraph features
  • all relation chains of length 3 involving this segment

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 20 / 24

slide-48
SLIDE 48

Model 3: Evidence graph (EG)

Segment feature sets:

  • base features incl. 2-node subgraph features:
  • position of the segment in the text
  • is it the first or the last segment?
  • has it incoming/outgoing edges?
  • number of incoming/outgoing edges
  • type of incoming/outgoing edges
  • 3-node subgraph features
  • all relation chains of length 2 involving this segment
  • 4-node subgraph features
  • all relation chains of length 3 involving this segment

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 20 / 24

slide-49
SLIDE 49

Model 3: Evidence graph (EG)

Segment-pair features:

  • direction of the potential link (forward or backward)
  • distance between the segments
  • whether there is an edge between the segments
  • type of the edge between the segments or None

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 21 / 24

slide-50
SLIDE 50

Results

scores reported as macro avg. F1

model cc ro fu at unknown

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 22 / 24

slide-51
SLIDE 51

Results

scores reported as macro avg. F1

model cc ro fu at unknown BL .861 .896 .338 .649

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 22 / 24

slide-52
SLIDE 52

Results

scores reported as macro avg. F1

model cc ro fu at unknown BL .861 .896 .338 .649 A-2 .578 .599 .314 .650 10.6% A-23 .787 .744 .398 .707 7.5% A-234 .797 .755 .416 .719 7.0% A-2345 .794 .762 .424 .721 6.8%

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 22 / 24

slide-53
SLIDE 53

Results

scores reported as macro avg. F1

model cc ro fu at unknown BL .861 .896 .338 .649 A-2 .578 .599 .314 .650 10.6% A-23 .787 .744 .398 .707 7.5% A-234 .797 .755 .416 .719 7.0% A-2345 .794 .762 .424 .721 6.8% A-2+r .861 .681 .385 .682 13.9% A-23+r .861 .783 .420 .716 11.3% A-234+r .861 .794 .434 .723 10.8% A-2345+r .861 .800 .443 .725 10.7%

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 22 / 24

slide-54
SLIDE 54

Results

scores reported as macro avg. F1

model cc ro fu at unknown BL .861 .896 .338 .649 A-2 .578 .599 .314 .650 10.6% A-23 .787 .744 .398 .707 7.5% A-234 .797 .755 .416 .719 7.0% A-2345 .794 .762 .424 .721 6.8% A-2+r .861 .681 .385 .682 13.9% A-23+r .861 .783 .420 .716 11.3% A-234+r .861 .794 .434 .723 10.8% A-2345+r .861 .800 .443 .725 10.7% EG-2 .918 .843 .522 .744 EG-23 .919 .869 .526 .755 EG-234 .918 .868 .530 .754

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 22 / 24

slide-55
SLIDE 55

Conclusions & Outlook

Conclusions:

  • first empirical study on the relationship between RST and ARG
  • majority of mappings canonical
  • tension between intentional and informational analysis in RST
  • automatically mapping RST to ARG
  • isomorphic structure mapping is not enough
  • EG model performes best

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 23 / 24

slide-56
SLIDE 56

Conclusions & Outlook

Conclusions:

  • first empirical study on the relationship between RST and ARG
  • majority of mappings canonical
  • tension between intentional and informational analysis in RST
  • automatically mapping RST to ARG
  • isomorphic structure mapping is not enough
  • EG model performes best

Outlook:

  • similar empirical analysis with longer text
  • try using RST parser output
  • augment arg mining text pipeline with RST features

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 23 / 24

slide-57
SLIDE 57

Literatur I

Nicholas Asher and Alex Lascarides. Logics of Conversation. Cambridge University Press, 2003.

  • Y. J. Chu and T. H. Liu. On the shortest arborescence of a directed graph. Science Sinica, 14:1396–1400, 1965.

Jack Edmonds. Optimum Branchings. Journal of Research of the National Bureau of Standards, 71B:233–240, 1967. James B. Freeman. Dialectics and the Macrostructure of Argument. Foris, Berlin, 1991. James B. Freeman. Argument Structure: Representation and Theory. Argumentation Library (18). Springer, 2011. William Mann and Sandra Thompson. Rhetorical structure theory: Towards a functional theory of text organization. TEXT, 8:243–281, 1988. Andreas Peldszus and Manfred Stede. From argument diagrams to automatic argument mining: A survey. International Journal of Cognitive Informatics and Natural Intelligence (IJCINI), 7(1):1–31, 2013. Andreas Peldszus and Manfred Stede. Joint prediction in mst-style discourse parsing for argumentation mining. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 938–948, Lisbon, Portugal, September 2015. Association for Computational Linguistics. URL http://aclweb.org/anthology/D15-1110. Andreas Peldszus and Manfred Stede. An annotated corpus of argumentative microtexts. In Argumentation and Reasoned Action: Proceedings of the 1st European Conference on Argumentation, Lisbon 2015 / Vol. 2, pages 801–816, London, 2016. College Publications. Manfred Stede. Handbuch Textannotation: Potsdamer Kommentarkorpus 2.0. Universitätsverlag Potsdam, 2016. Manfred Stede, Stergos Afantenos, Andreas Peldszus, Nicholas Asher, and Jérémy Perret. Parallel discourse annotations on a corpus of short texts. In Proceedings of the International Conference on Language Resources and Evaluation (LREC), Portoroz, 2016.

Peldszus, Stede (Uni Potsdam) Rhetorical structure and argumentation structure ArgMin WS 3 24 / 24