Commit
·
bc8b5d9
1
Parent(s):
2fa3d6f
add dataset and format description and
Browse files
README.md
CHANGED
@@ -1,3 +1,98 @@
|
|
1 |
---
|
2 |
license: cc-by-nc-sa-4.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: cc-by-nc-sa-4.0
|
3 |
---
|
4 |
+
# An annotated corpus of argumentative microtexts
|
5 |
+
|
6 |
+
The arg-microtexts corpus features 112 short argumentative texts. All texts
|
7 |
+
were originally written in German and have been professionally translated to
|
8 |
+
English.
|
9 |
+
|
10 |
+
The texts with ids b001-b064 and k001-k031 have been collected in a controlled
|
11 |
+
text generation experiment from 23 subjects discussing various controversial
|
12 |
+
issues from [a fixed list](topics_triggers.md).
|
13 |
+
|
14 |
+
The texts with ids d01-d23 have been written by Andreas Peldszus and were
|
15 |
+
used mainly in teaching and testing students argumentative analysis.
|
16 |
+
|
17 |
+
All texts are annotated with argumentation structures, following the scheme
|
18 |
+
proposed in Peldszus & Stede (2013). For inter-annotator-agreement scores see
|
19 |
+
Peldszus (2014). The (German) annotation guidelines are published in Peldszus, Warzecha, Stede (2016).
|
20 |
+
|
21 |
+
The texts with ids b001-b064 and k001-k031 have been collected in a controlled
|
22 |
+
text generation experiment from 23 subjects discussing various controversial
|
23 |
+
issues. Below is a list of the topics and the corresponding trigger question in the experiment.
|
24 |
+
|
25 |
+
Note that the subjects in the experiment chose from the German trigger questions and wrote German text, that was only later professionally translated to English. For more information about the experiment and the translation, see ECA Paper.
|
26 |
+
|
27 |
+
## DATA FROMAT (ARGUMENTATION GRAPH)
|
28 |
+
|
29 |
+
This specifies the argumentation graphs following the
|
30 |
+
annotation scheme described in
|
31 |
+
|
32 |
+
Andreas Peldszus and Manfred Stede. From argument diagrams to argumentation
|
33 |
+
mining in texts: a survey. International Journal of Cognitive Informatics
|
34 |
+
and Natural Intelligence (IJCINI), 7(1):1–31, 2013.
|
35 |
+
|
36 |
+
|
37 |
+
An argumentation graph is a directed graph spanning over text segments. The
|
38 |
+
format distinguishes three different sorts of nodes: EDUs, ADUs & EDU-joints.
|
39 |
+
|
40 |
+
- EDU: elementary discourse units
|
41 |
+
The text is segmented into elementary discourse units, typically at a
|
42 |
+
clause/sentence level. This segmentation can be the result of manually
|
43 |
+
annotation or of automatic discourse segmenters.
|
44 |
+
|
45 |
+
- ADU: argumentative discourse units
|
46 |
+
Not every EDU is relevant in an argumentation. Also, the same claim might
|
47 |
+
be stated multiple times in longer texts. An argumentative discourse unit
|
48 |
+
represents a claim that stands for itself and is argumentatively relevant.
|
49 |
+
It is thus grounded in one or more EDUs. EDU and ADUs are connected by
|
50 |
+
segmentation edges. ADUs are associated with a dialectic role: They are
|
51 |
+
either proponent or opponent nodes.
|
52 |
+
|
53 |
+
- JOINT: a joint of two or more adjacent elementary discourse units
|
54 |
+
When two adjacent EDUs are argumentatively relevant only when taken
|
55 |
+
together, these EDUs are first connected with one joint EDU node by
|
56 |
+
segmentation edges and then this joint node is connected to a corresponding
|
57 |
+
ADU.
|
58 |
+
|
59 |
+
### edge type
|
60 |
+
The edges representing arguments are those that connect ADUs. The scheme
|
61 |
+
distinguishes between supporting and attacking relations. Supporting
|
62 |
+
relations are normal support and support by example. Attacking relations are
|
63 |
+
rebutting attacks (directed against another node, challenging the accept-
|
64 |
+
ability of the corresponding claim) and undercutting attacks (directed
|
65 |
+
against another relation, challenging the argumentative inference from the
|
66 |
+
source to the target of the relation). Finally, additional premises of
|
67 |
+
relations with more than one premise are represented by additional source
|
68 |
+
relations.
|
69 |
+
|
70 |
+
Values:
|
71 |
+
- seg: segmentation edges (EDU->ADU, EDU->JOINT, JOINT->ADU)
|
72 |
+
- sup: support (ADU->ADU)
|
73 |
+
- exa: support by example (ADU->ADU)
|
74 |
+
- add: additional source, for combined/convergent arguments with multiple premises (ADU->ADU)
|
75 |
+
- reb: rebutting attack (ADU->ADU)
|
76 |
+
- und: undercutting attack (ADU->Edge)
|
77 |
+
|
78 |
+
### adu type
|
79 |
+
The argumentation can be thought of as a dialectical exchange between the
|
80 |
+
role of the proponent (who is presenting and defending the central claim)
|
81 |
+
and the role of the opponent (who is critically challenging the proponents
|
82 |
+
claims). Each ADU is thus associated with one of these dialectic roles.
|
83 |
+
|
84 |
+
Values:
|
85 |
+
- pro: proponent
|
86 |
+
- opp: opponent
|
87 |
+
|
88 |
+
### stance type
|
89 |
+
Annotated texts typically discuss a controversial topic, i.e. an issue posed
|
90 |
+
as a yes/no question. Example: "Should we make use of capital punishment?"
|
91 |
+
The stance type specifies, which stance the author of this text takes
|
92 |
+
towards this issue.
|
93 |
+
|
94 |
+
Values:
|
95 |
+
- pro: yes, in favour of the proposed issue
|
96 |
+
- con: no, against the proposed issue
|
97 |
+
- unclear: the position of the author is unclear
|
98 |
+
- UNDEFINED
|