Corpus of GraphicalTextual Presentations
Giuseppe Carenini  97
Motivations
What
this corpus is not
Sources
The Corpus is organized in categories that are not mutually exclusive
or complete (i.e some presentations belong to more than one category and
you may easily discover new presentations that do not fit in any category).
As soon as the work on presentational goals will reach a stable state,
the corpus will be reorganized, probably the "argumentation categories"
will be restructured to include most of the presentations.
If you find anything wrong, confusing, interesting etc. please send
your comments to carenini@cs.ubc.ca
. A sample presentation is uniquely identified by the last number in its
identifier. We are also interested in any new presentation you might find
that can be added to the corpus.
For AB group (PittCMU): * indicates that the design might not
be possible in xSAGE (please send me comments about this judgement).
Argumentation

Nothing has changed it... so nothing will (
ex1 5)

Highest than ever, yet low.... (
ex16)

More of X does not affect Y...(
ex17?*? )

Argumentative annotation (
ex18)

Causal explanation for change (or nochange)(
ex22,
ex3 3,
ex44*,
ex11 ). One of the communicative function of a chart is to show how
an eventualty (event, process, state) affected a variable. A simple case
is the beforeafter chart in timeseries.
We can envision more complex cases for all the possible combinations of
temporal relations between eventualities and variables ( During X, Y
increased  Every time X, Y decreases  For the first time ...  ...).

Violated expectations. Something changed without plausible reason.(
ex133* )

Unplausible correlation: whenever plotting some data indicates an "impossible"
correlation, the chart can be used as evidence of "something wrong" in
the source of the data (ex134).
This picture is not local sometimes it takes a while to unload.

X is better than Y...( ex19*
, ex210*
)... X is an useful index (ex111*)

Complex argument...( ex112
, ex213
*)
Interesting graphic designs

Bar chart with absolute value and label with percentage (ex114
)
or opposite ( ex530).
Very common in the Economist.

Subset is compared with superset respect to some attributes (
ex115). You can obtain such a presentation in Envisage by selecting
the subset. Obviously, these graphs, by pointing out a specific subset
of a set, can play very different argumentative functions. I discuss such
differences in my document on communicative goals.

Aggregation (again, the selection of particular aggregations disaggregations
can have different argumentative functions...)

Mix of aggregates and nonaggregates (or aggregates at different levels)
in the same chart (
ex116,
ex217,
ex318, ex419
, ex520*)

Mix of aggregates and nonaggregates in different charts in the same presentation
(
ex12 )

Different ranges in different charts for the same presentation (
ex1 2)

Aggregates are expicitly removed from data (ex10*)

Show conclusion and raw data(ex122*)

Meaningful paths and areas in relational spaces (i.e. no maps) . I believe
this class of graphs should be carefully studied (ex123*,ex224*,ex325*).
Interesting combination of graphic and text

Text discusses subsets (including one element only) with particular interesting
features (
ex126, ex227,
ex328
,
ex530,
)
. More detailed analysis in my document on communicative goals.

Aggregation/Disaggregation (
ex131,
ex232
,
ex319)
Maps (very preliminary)
Graphical annotations (more to be included,
Vibhu is also working on this now)

One graph is the expansion of a part of another graph (
ex143* ). Source: Kosslyn

On the axes (
ex133*  F'cast on Xaxis). I suggested to use the same technique
to identify shortfalls in the IUI paper.

Extremely large values (ex114)

Areas between two lines (
ex18
)

Aggregation (
ex132 )
Miscellaneous

Present only a subset to make a point. This is probably true for many charts,
but it is difficult to identify because usually the whole data set is not
mentioned (
ex134 )

"A graph that provides more info than the readers need forces them to filter
and search. Directly label only the critical values."(
ex135). Source: Kosslyn
to do: all the ones form Economics (E)