The past is an unattainable country. All access to it is mediated by myriad artifacts. To make sense of this morass of information, historians invented concepts (such as “the French Revolution” or “the Dutch Golden Age”) that bind an otherwise unconnected set of entities together. These historical periods, however, remain constructions and their meaning a moving target. In this paper, we outline a search interface that allows researchers to explore mentions to past events or periods in digital heritage. We show how semantically-enhanced search enables users to retrieve information related to complex concepts. After introducing the general architecture and the interface, we showcase it by elaborating on one pilot study, targeting the Golden Age in contemporary Dutch parliamentary discourse.
The past is not a foreign, but an unattainable country. All access to it
is mediated via myriad artifacts, which historians attempt to weave
together in convincing narratives [
From an Information Retrieval perspective, this amounts to a
daunting task: finding documents that refer to fragmented and
heterogeneous concepts is fraught with multiple methodological and
philosophical hurdles. How, for example, can a machine retrieve
all documents related to the “French Revolution”? To contextualize
the problem, Shaw [
SEMANTiCS 2017 workshop proceedings: EVENTS
September 11-14, 2017, Amsterdam, Netherlands
to Information Science. The notion of colligation originated from
Whewell [
The remainder of the paper is structured as follows: Firstly, we describe WideNet, a novel (re)search interface that is designed to explore historical periods in diachronic corpora. Secondly, we discuss a specific pilot study—the Golden Age in contemporary Dutch Parliamentary discourse—to demonstrate how semanticallyenhanced search supports historians in studying how the past is remembered. 2
WideNet builds on a semantically enriched version of the Dutch
parliamentary proceedings: the “verbatim” record of all debates in
the Staten Generaal). These discussions touch on almost every issue
that moved Dutch public opinion during the last two centuries.
Despite its centrality in the political landscape, the proceedings’
unwieldy size made it dificult for historians to explore. The
existing index is rather limited in scope, which makes searching for
infrequent and complex items a laborious, if not impossible, task.
Digitization and enrichment have unlocked this resource in novel
ways. WideNet, we believe, is a valuable addition to this trend, since
it accommodates a growing need for complex search systems in
the Digital Humanities. The application of semantically-enhanced
search to large digitized text collections was first proposed by Hinze
et al. [
To prepare the corpus for semantically-enhanced search, all the
documents were processed by a semantic annotation system, which
linked concepts and entities in the text to a Knowledge Base (KB)
[
To map abstract concepts like “Dutch Golden Age” to more
specific concepts and entities, we extracted a subgraph of DBpedia
into a property graph database (see [
The user-interface guides scholars through three phases: (1)
selection of the root category, (2) assessment of the categories’ and
entities’ relevance (3) close-reading of selected documents. In the
ifrst step, the user selects a root category from a typeahead search
box (see Figure 1), and demarcates the query by selecting a time
1For namespace prefixes, see https://dbpedia.org/sparql?nsdecl.
period, which is used to prune the underlying entities of the
selected categories. WideNet then retrieves the network of narrower
categories for each selected root category and collects the
contained entities as potentially relevant query components. Behind
the scenes, each entity is compared with the target period (i.e. time
interval), and is considered to be outright relevant to the period, or
not, or a borderline case, or as lacking temporal clues altogether.
In the current implementation this classification is achieved with
simple rules, based on the features: ‘fraction of years within period,’
‘fraction of intervals that overlap with the period,’ and ‘has at least
one year in period.’ The system uses chronological information to
deselect (sub)categories where more than half of the dated member
entities are out-of-period. Subsequently, the user assesses which
of the retrieved subcategories actually contain entities that lead
to relevant results (see Figure 2). This step was motivated by the
following observations: Firstly: many of entities returned by
Spotlight were simply incorrect and therefore should be discarded easily.
Secondly: as we lack a clear ontology of our target—as stated earlier,
colligatory concepts are by definition dependent on the perspective
of the user—the interface shows a wide range of potentially
relevant entities, but defers the actual selection to the scholar, who is
ultimately responsible for judging the “aboutness” of a reference
[
The study of “memory”—the diverse ways through which historical
events reverberate over time—has attracted the interest of
historians and other scholars working in the humanities[
Traditionally, scholars in the humanities tend to base their
arguments on a small set of carefully selected and curated material.
Instead of building conclusions on small datasets, the digital turn
has enabled historians to “holistically assess the typicality, scope,
and power of key issues” [
We started by sifting through all the entities WideNet returned as related to the Golden Age, using the interface shown in Figure 2.3 Having a thematic overview of all entities facilitated this task for three reasons. Firstly: as entity-linking remains an error-prone process, many of the items found by WideNet turned out to be irrelevant. For example the historical person of Michiel de Ruyter was often confused with a hospital named after him. Judging whether an entity relates to the topic at hand is, as we argued previously, 2see http://politicalmashup.nl/ and https://dilipad.history.ac.uk/ 3The original data are available on https://widenet.e.hum.uva.nl/preview/ge/ ultimately vested in the historian who models the period under investigation. In line with the description of colligatory concepts, the design of the WideNet interface facilitates this task by actively engaging the user to refine the search results and define the target query.
Secondly: as the overview groups all entities by their subcategory, we can easily identify the distinct topics and spheres of identity construction. We labeled the found entities as either belonging to the economical (i.e. Dutch East India Company), cultural (i.e. Spinoza), or the political sphere (i.e. Eighty Years’ War). The coarse-grained subcategory-wise exploration of the retrieved entities foregrounded the diversity of the search result. It, thereby, also made long-tail entities clearly visible: grouping entities by subcategory prohibits the rare ones from drowning in a morass of highly frequent (but potentially irrelevant) results. For example: mentions of the architectural artifacts of the seventeenth-century appeared very often, but did not co-occur with any relevant event or person (and were thus conveniently discarded with one click), while references to the Eighty Years’ War were scarce but highly informative.
Thirdly: besides thematically grouping the references, WideNet allows the user to aggregate the selected entities by their metadata ifelds. For our purposes, we were particularly interested in the distribution of these mentions over time and party. From a temporal perspective, the number of Golden Age references showed a slight increase after 2002, but not drastically (result not reproduced here). Aggregating the found references by party revealed deeper-rooted discrepancies in the engagement with the past. The discourse of the conservative and populist right showed a greater tendency to invigorate their nationalistic appeals with historical references. As shown in Figure 4 the Reformed Party and the Party for Freedom (PVV), were proportionally the most active in this respect, followed by the right-liberal People’s Party for Freedom and Democracy (VVD). The fact that most of these mentions stemmed from the right, suggests that the Golden Age functioned as an instrument for forging a more exclusive, culturalist understanding of Dutchness. This applied even more to political references, which almost exclusively circulated among those seated at the right. Mentions of economic entities showed a more balanced distribution, skewing even a bit towards the left (results not reproduced here).
By aggregating the selected speeches, WideNet enables researchers to conveniently map their data along diferent axes. But what do these aggregated diferences actually mean? How, exactly, do these speeches enact of identities? What are the specific linguistic instruments that “bound and bond people” into distinctive groups? To answer these questions, we studied the concrete, fine-grained, mechanics at play in the speeches that survived the filtering process (as shown in Figure 3). Linguistically, the use of personal pronouns, such as “we” became apparent, especially in the in the context of political entities—which mainly comprised wars and taxation, such as the Eighty Years’ War and the “tiende penning,” a VAT on movables, introduced by the Duke of Alva. For example, Verheijen a MP of the right-liberal VVD, evoked these events when commenting on proposals which would give Europe more say in national taxation matters. To support his argument he asserted that “we waged 80 years of war against Spain to obtain our independence.” In this passage, Verheijen inserts a transhistorical “we” in his speech, he amalgamates the Dutch who fought Spain with those of today, who still fiercely resist infringement on their national independence.
Generally, a fine-grained linguistic analysis, demonstrated that MPs on the right exhibited greater intimacy with the past: more often did they invoke the “we” as a homogeneous national actor, or used cognitive verbs to suggest direct access to the minds of illustrious Dutchmen from the past. Another example is Madlener of Wilders’ PVV, who insisted that “we won against Spain”—again a reference to the Eighty Years’ War during the seventeenth century. Moreover, he exclaims: “I think the Sea Beggars are deeply ashamed of your remark”, thereby rejecting the claims of an opponent by straightforwardly probing and exposing the minds of actors from the distant past.
While the memory of political events was dominated by the right, the discussion about the economic ramifications of the Golden Age was more evenly distributed among left and right—but nonetheless contested. The interpretation of the Dutch colonial heritage and its trade practices figured here as the main bone of contention. This debate was largely sparked by a remark of then-prime-minister Jan Peter Balkenende, who urged the Dutch to embrace their “VOCmentality,” which he characterized as a tradition of risk-taking and brave, global entrepreneurship. The depiction of the VOC as the lfagship of Dutch capitalism—based on the values “Freedom, Entrepreneurship, and Competition”, according to Ten Broeke of the VVD—bounced against a wall of skepticism raised by left-wing MPs, who mostly emphasized the exploitative practices of the colonial past. Vendrik of GreenLeft, for example, posits that the VOC mentality actually means “becoming rich at the expense of others”, while Irrgang, a member of the Socialist Party, equates the VOC— and the underlying mentality—with “nothing more than colonial plundering, the creation of monopolies.”
The Golden Age serves just as one example of how WideNet supports the scholar to study historical periods in large corpora. We have shown how the WideNet interface enables historians to explore colligatory concepts in diferent stages: First, the user selects the query from an existing Knowledge Base, after which WideNet gathers all documents that contain references to the diferent aspects of the historical period under investigation. The user is then presented with an overview of the found subcategories and their associated entities. At this stage the user decides which of the found events and persons are relevant and should be retained and subjected to close reading. Eventually, the findings can be aggregated and plotted along various dimension such as time or party.
During the search process, WideNet logs all activity and thereby collects a fair amount of data. Besides the initially selected target categories and time period, it stores each decision about the relevance of the entities, as well the individual documents. In other words, WideNet gathers how users model historical events. These data are expected to be valuable for two purposes. Firstly, as a gold standard for the (temporal) pruning of raw category trees. Collecting user data enables the system to automatically deselect irrelevant entities (which would still be reversible by the user). Secondly, logging the decisions made by many users might foreground academic or societal disputes about certain events in the past. Besides providing a useful front-end for historical research, the interface’s back-end could return valuable data for humanities scholars, as it may expose diferent understandings of history that circulate in society.