Integrating Tools for Synthesis into Digital Libraries
Ryan Shaw, Michael Buckland and Ray Larson
School of Information
University of California, Berkeley
Berkeley, California, 94720
{ryanshaw, buckland, ray}@ischool.berkeley.edu
ABSTRACT
Second-generation digital libraries aim to go beyond pro-
viding access to resources, toward integrating tools and ser-
vices for exploring the content of resources. Much of the
work in this area focuses on tools for analyzing data sets
such as numeric observations or text corpora. Our work,
in contrast, focuses on tools for synthesizing an understand-
ing of mentioned but unexplained names or terms. We argue
that these self-service reference tools, by making connections
across separate collections, complement analytic tools that
uncover patterns within a collection. Finally, we demon-
strate working prototypes that illustrate the basic principles
of self-service reference.
1. ANALYSIS AND SYNTHESIS
Research into integrating digital library content with com-
putational tools and services has been primarily concerned
with examining, analyzing, and finding patterns within a
digital data set. A digital data set is ordinarily enhanced
by adding formal elements to aid analysis. These elements
allow researchers to identify patterns within a data set and
to navigate among these patterns and structures. Such ele-
ments include markup, such as TEI, or features designed for
statistical machine learning, data-mining, and other tech-
niques. With markup, the patterns are explicitly revealed
in databases representing structures of relationships, while
statistical techniques uncover latent patterns among the de-
fined features.
Traditional library work, however, tends to focus on synthe-
sis. Some work is concerned with analyzing individual data
sets, as when cataloging a collection or compiling a bibliog-
raphy, but the focus on a data set is qualified by an effort to
make what is inside the data set interoperable with mate-
rials elsewhere through standardized formats and standard
descriptions. Other library work, notably reference work
and much bibliographical work, looks outwards to find re-
lated material elsewhere.
Like the elements added to enable analysis, the markup and
metadata added by librarians not only describe documents
but also form structures for finding documents, as in infor-
mation retrieval, or for discerning patterns within a popula-
tion of documents, as in bibliometrics. Some useful kinds of
metadata (such as titles and bibliographical citations) are
not added by librarians but can be usefully exploited by
them, for example, expanding subject access with keywords
derived from titles and creating citation indexes to identify
related prior work, co-citing research fronts, and interdisci-
plinary connections.
Scholarship involves an interplay between analytic and syn-
thetic modes, between representing the structure of data
and building structures to connect data [5]. There is a large
opportunity for linking analysis-oriented tools and services
with the synthesis-oriented tendencies of library and biblio-
graphical efforts. In both cases markup and metadata play
dual roles as descriptions and also as facilitating infrastruc-
ture [2].
A major barrier to closing this gap is that researchers typi-
cally focus their structure-building efforts on individual data
sets, while librarians strive to create structures that traverse
a wider documentary universe. Researchers might focus on
producing a data set that is richly structured internally, yet
relatively disconnected from the external universe of docu-
ments. Librarians, on the other hand, tend to treat such a
data set as a unit, adding some descriptive metadata to give
it a place within the larger set of resources, but failing to ex-
ploit the richness of its internal structure. Changing this sit-
uation requires better practices on both sides. Researchers
need to think more about how the elements they produce
will participate in the world outside the scope of their cur-
rent project, while librarians need to be more aware of the
internal structure of the resources they organize. Both need
to appreciate the potential benefits.
2. SELF-SERVICE REFERENCE
Our current research is based on the principle that the differ-
ence between seeing and understanding lies in knowing the
context and relationships of whatever is of interest. This im-
plies that identifying relationships between items in a data
set is important, but understanding also depends on being
able to relate any item to resources in the wider environment
outside the data set capable of providing explanations and
relationships.

Figure 1: Interface to an RDF data set of events in Emma Goldman’s life.
Reference libraries can be very good resources for acquiring
some knowledge of the context and relationships of things.
Through its selection and organization of resources, the ref-
erence library enables patrons to conduct initial investiga-
tions and orient themselves to unfamiliar topics. Yet re-
search into reference service in a digital environment has
mainly focused on librarians answering questions for pa-
trons, rather than on the design of services for efficient and
effective self-help [4].
Ideal self-service reference tools do not merely list resources,
but empower users to select the best specific resources or
fragments of resources to achieve their ends. Such tools
need to perform a number of related functions. First, the
tools need to indicate to users which reference resources are
relatively trustworthy. Then, for any specific query, they
need to indicate which resources are likely to yield useful
results. In order for these results to materialize, the tools
must address the differences in descriptive vocabulary be-
tween different resources. They also need to facilitate the
comparison of results from several resources, so that users
can verify and cross-check the explanations they receive. Fi-
nally, they must be very easy to use, or they won’t be used.
3. PROTOTYPES FOR DEMONSTRATION
We are building tools for enhancing digital documents in the
humanities domain by identifying references to people, or-
ganizations, places, topics, or events, disambiguating these
references by linking them to identifiers from naming au-
thorities, and using the disambiguated references to provide
links to related explanatory resources available on the web.
Distinguishing referenced entities by facet greatly facilitates
search. Standard natural language processing techniques
can analyze the text to identify entity references, and the
reader can also use a web browser-based annotation tool.
Our designs are being developed in two parallel projects:
one “Bringing Lives to Light: Biography in Context”1 uses
the condensed texts of biographical dictionaries; the other,
“Context and Relationships: Ireland and Irish Studies”2 uses
digitized back runs of journals on Irish culture and history.
Initially we developed three “proof of concept” prototypes
each focused on a different requirement. The first, based on
a chronology of Emma Goldman’s lecture tours, looks into
a data set, adding markup to support geo-temporal naviga-
tion to identify for individual speeches illustrative materi-
als within the Emma Goldman Papers. The Emma Gold-
man Papers editors, like many editors of historical papers,
maintain a day-by-day chronology detailing where Emma
Goldman and her associates were and what they were doing.
This chronology serves as an internal reference tool for the
project, allowing the editors to make inferences about when
or where documents may have been produced and to check
for inconsistencies in historical accounts. But as it runs to
thousands of entries, it is not available in the printed vol-
umes, despite the fact that it could be a valuable reference
for Emma Goldman scholars outside the project as well.
1http://ecai.org/imls2006/
2http://ecai.org/neh2007/

Figure 2: Integrating external resources into a digital library of full-text documents.
Starting with a text document containing the chronology
for the years 1910 through 1916, we produced an RDF data
set by parsing dates, geocoding place names, and disam-
biguating personal names by linking them to DBpedia. We
then scanned a selection of documents from the Emma Gold-
man Papers’ archives, put them online, and manually linked
these scans with relevant events from the chronology. Fi-
nally, we developed a browser-based interface for exploring
the RDF data set (Figure 1). All the data extracted from
the chronology is encoded in the interface’s HTML using
RDFa [1], and all the querying is done locally. Essentially
it is a self-contained, human- and machine-usable data set
of information about events in Emma Goldman’s life. Yet
because the information is embedded using the RDFa stan-
dard, we can easily extract the RDF triples from the HTML
page and do something else with it, such as merge it with an-
other data set or use SPARQL to query it in ways that aren’t
possible through this interface. From the Emma Goldman
Papers’ point of view this approach is very low-maintenance:
since this is a static HTML file it just needs to be placed on
a web server; there is no need to maintain a database or a
dynamic web application framework.
The second prototype (Figure 2), using pages of Irish pe-
riodicals and modeled on how reference works are used, is
outward-looking, supporting real-time queries to appropri-
ate specialized external resources whenever the reader de-
sires an explanation of a personal name, place name, topic,
or event mentioned in the text. The goal is to demonstrate
how external resources might be made more readily avail-
able in a digital library of full-text documents. From our
partners at Queen’s University Belfast we have a collection
of approximately 50,000 scanned pages of journal articles,
from the 1780s to the present, focusing on Ireland. To these
we have added the full text of 100 books on Irish history and
culture from the Internet Archive’s book-scanning project.
We aim to integrate three kinds of functionality into this
library, which we call context finding, context building, and
context providing. Context finding means a reader should
be able to move from a name in a text to resources that
describe, explain, or otherwise provide resources for under-
standing that name. One might think of it as moving from
names in texts to entries in reference works that explain
those names. Context building means the ability to add
links to such resources to the texts. In our systems context
building is done both manually (via annotation) and algo-
rithmically (using named-entity detection). Finally, context
providing means aggregating across a corpus of texts all the
references to a particular name. This is an inversion of con-
text finding, and one might think of it as a union index that
supports moving from an name in a reference work to a list
of locations in texts where that name is used.
Searches depend on unambiguous queries, so the third pro-
totype, applied to biographical articles in Wikipedia and
similar sources, focuses on invoking established naming au-
thorities to disambiguate personal names (e.g. Which King
Charles?). To resolve ambiguities a simple interface links to
the appropriate authoritative identifiers. For names of per-
sons and organizations, these include Library of Congress
and Deutsche Nationalbibliothek authority files and World-

Cat Identities URIs. For place names we use identifiers
from the GeoNames geographical database. We are using
Wikipedia and Freebase URIs to identify historical events
for which no established naming authority currently exists.
The system is extensible: new sources of identifiers can be
added.
After entities have been unambiguously linked to authorita-
tive identifiers, these identifiers are used to find additional
information about them. For example, a linked map and
timeline displayed alongside a document may indicate the
locations of places and the locations and times of events
referenced in the document. Access to unstructured infor-
mation and related documents can be provided by using
identifiers or aliases to construct dynamic links represent-
ing queries on appropriate reference sources or special col-
lections. We are currently working on representing all this
auxiliary information—disambiguated names and relation-
ships, factoids, and links to relevant resources—using the
same Linked Data3 approach taken in the Emma Goldman
prototype. This will enable us to combine the functionality
of all three prototypes in future versions.
4. FACILITATING SYNTHESIS
Identifying and disambiguating references to entities within
data sets, so as to represent the interrelationships among
these various entities and thus provide richer context for in-
terpretation, is common practice in computational research
projects. Typically such projects mint their own identifiers
for these entities, which is sufficient for identifying entities
within the scope of that project and within that data set, but
fails to support their use to create bridging links to other,
external resources concerning those entities. Use of shared
naming authority services would go a long way towards ame-
liorating this situation. Librarians have deep experience
building such services, yet matching queries to unfamiliar
vocabularies on external network-accessible resources has re-
ceived relatively little attention [6, 3].
In the humanities, place name gazetteers, encyclopedias, bi-
ographical directories, dictionaries of concepts, and other
long-established reference resources are services for finding
and providing contextual background. The elements needed
for such services, including identifiers for entities and con-
cepts of interest and various kinds of semantic relationships
for linking them to one another, are being produced by dig-
ital humanities scholars, but they aren’t yet being aggre-
gated into coherent or interoperable wholes. If they were,
the fruits of humanities scholarship would be accessible to
a wider audience and opened to wider participation in such
scholarship. Generating such services must yield tangible
benefits for scholars, without straitjacketing them.
The affordances of the classic reference tools of the humani-
ties have been dominated by the structure of the codex and
the high cost of printing. Neither constraint applies in a
digital environment. As a result a full agenda emerges for
systematically redesigning traditional tools for a new and
different environment and reconstructing the amenity of a
reference collection in the digital library environment [4].
3http://en.wikipedia.org/wiki/Linked Data
5. CONCLUSIONS
Understanding any datum depends on knowing the context
and background. For this reason alone, the development of
better tools for linking any datum to explanatory resources
outside as well as inside the data set of which it forms a
part is important for both research and education. Further,
doing this has the potential for repositioning how we under-
stand the data set itself in relation to a wider context and,
thereby, also suggesting relationships between data sets that
are unlikely to be evident so long as the focus remains on
what is inside the data set.
6. ACKNOWLEDGMENTS
The work discussed in this paper has been generously funded
by the Institute of Museum and Library Services and the
National Endowment for the Humanities.
7. REFERENCES
[1] B. Adida and M. Birbeck. RDFa primer. W3C working
draft, W3C, June 2008.
http://www.w3.org/TR/2008/WD-xhtml-rdfa-primer-
20080620/.
[2] M. Buckland. Description and search: Metadata as
infrastructure. Brazilian Journal of Information
Science, 0, 2006.
[3] M. Buckland, A. Chen, H. Chen, Y. Kim, B. Lam,
R. Larson, B. Norgard, and J. Purat. Mapping entry
vocabulary to unfamiliar metadata vocabularies.
Technical report, Corporation for National Research
Initiatives, 1999.
[4] M. K. Buckland. Reference library service in the digital
environment. Library & Information Science Research,
30(2):81–85, June 2008.
[5] W. McCarty. What’s going on? Literary & Linguistic
Computing, 23(3):253–261, Sept. 2008.
[6] C. Plaunt and B. A. Norgard. An association-based
method for automatic indexing with a controlled
vocabulary. Journal of the American Society for
Information Science, 49(10):888–902, 1998.