• Generate a message about a new (not previously used) possible foraging behavior, if any deduced; unusual (historically infrequent) behaviors are selected over common ones.

Remove redundancy. Aggregate the messages generated for the week through these two steps to remove redundancy (such as by grouping together days with similar behaviors).

Paragraph 3

Movement pattern. Generate a message for a question or comment based on the movement pattern—C1, C2 or C3— with the aim of intriguing the reader.

Micro planner and sentence realiser. The micro planner takes the messages generated by the document planner, implements aggregation through a variety of linguistic devices (such as ranges, coordination, and subordination), and limits linguistic repetition by varying the vocabulary. It provides sentence specifications to the “sentence realiser,” which then generates sentences using the SimpleNLG library. 6

Figure 4 is a screenshot of the Blog-
ging Birds interface in which an auto-
matically generated weekly blog for a
kite is overlaid on a Google map of the
bird’s whereabouts with its historical
home ranges marked as blue poly-
gons. In this example, Wyvis, one of
five red kites being blogged about, has
traveled between two home ranges
(movement pattern C3), and an expla-
nation for the observed movement
pattern is provided based on the age
The content is selected through a
process of summarization and aggre-
gation of information. This is the sec-
ond creative aspect of the blog genera-
tion (the first involved imagining a
wide range of possible behaviors), as it
plans what story to tell from the imag-
ined behaviors. Blogging Birds aims to
provide an overview of the main behav-
iors and highlight aspects that might
be interesting to the human reader.
Movement behaviors are considered
more interesting than foraging behav-
iors, and rarer foraging behaviors are
prioritized over more frequent ones.
Each blog attempts to inform the read-
er about different aspects of red kite
ecology by selecting different behav-
iors from different days. The main
steps are as follows:

Paragraph 1

Movement pattern. Generate a message based on the detected movement pattern—C1, C2, or C3 in Figure 2; if the age of the bird can be used to interpret this pattern, add such an interpretation message;

Habitats visited. Generate a message summarizing the habitats visited; and

Other kites. Generate a message about other kites recorded nearby, if any.

Paragraph 2

Days of the week. Iterate over each day of the week (Monday to Friday):

•If the bird remained relatively
static—C1 in Figure 2—then generate
a message about nearby places or gen-
erate a message about any movement
behavior detected; and
particular patterns are detected in the
data. In total, the system implements
Drools for 26 movement behaviors
(such as flying along a coast or over a
landmark like a castle or loch and the
home-range-related movement pat-
terns in Figure 2); 33 foraging behav-
iors, mostly detailing the food avail-
able for a kite in different habitats at
different times of the year but also
sometimes related to specific features
(such as when a red kite near a road
might be looking for roadkill); and six
social behaviors (such as roosting and
nesting); see the online appendix “Ex-
ample Rules” ( dl.acm.org/citation.cfm
?doid=3231588&picked=formats).

The pattern-detection module then exhaustively applies the rules to the satellite fixes to produce a list of all observed movement behaviors and all possible foraging and social behaviors consistent with known environmental and geographic conditions. The latter is the first step in the creative process, whereby the program explores the conceptual space to “imagine” how the kite might have been behaving.

Blogging Birds uses a rule-based text planner for dynamic text generation. The planning rules decide how information is ordered, but what information to include and how to organize it into sentences is determined at run-time in a data-driven manner.

The blogs are always planned as three paragraphs, the first describing the overall trends, the second providing more detail on a day-to-day basis, and the third posing a question about what the kite might do next, as well as occasionally offering a conclusion.

Table 2. Baseline computer-generated blog without reference to ecological concepts for the week outlined in Figure 4.

Wyvis had enough of the area around Teavarran and decided to move to Crieff approximately 73 miles away. No doubt Wyvis had a social week, as kites Moray and Millie were often seen in the vicinity.

Monday, Wyvis spent most of her time around Torness, Errogie, and Teavarran. On Tuesday evening, she reached moorland near Crieff flying approximately 65 miles amid cloudy conditions and averaging a remarkable 11 miles per hour. The next five days Wyvis spent most of her time around Edinample, Tullybannocher, and St Fillans. During this time, she was seen mainly on acid grassland while making occasional journeys to farmland.

Will Wyvis settle down here?

Figure 4. Screenshot of the Blogging Birds Web interface.