The Estado da ´India constituted the most complex overseas Portuguese set of territories. This paper investigate a digital methodology employing Natural Language Processing to study historical events regarding it during the period 1616-1618. We explore the application of an event extraction tool over an extract of the The Monsoon Books. Our preliminary results expose the current problems and help us shape further work for automatic processing of historic corpora.
A process like this applied to the study of a colonial macro-region as the eastern Portuguese Empire identifies and categorize human actions and episodes which determine not only patterns of historical junctures in time and place, but also disruptive events that underline changing processes. 2
The Documentos Remetidos da ´India or Livros das Monc¸˜oes (Monsoon books)
collect letters exchanged between the monarchs and Portuguese government
councils and India viceroys where all types of afairs concerning the so-called
Portuguese Estado da I´ndia were discussed. They comprise a geographical scope
from Eastern Africa to Japan. The use of this collection is paramount to
understand the internal dynamics of the Portuguese Estado da ´India until the 19th
century. In fact, they are considered the core documents produced by Portuguese
authorities in Asia. The fact of being a type of documental corpora concerning
all types of issues makes the Monsoon Books unique and a privileged lab for
building a new analytic model and approach to understand internal dynamics of
colonial empires macro-regions. The Monsoon Books are composed by the sets of
documents located in both in the Portuguese National Archives, in Lisbon, and
in the Historical Archives of Goa, in Panjin, India. Since this paper intends to
test an automatic event extraction model in order to conceive an interpretative
framework of European colonial presence in overseas macro-regions, we employ
some of the already transcribed and published books referring to the years of
1616-1618 [
The goal of event identification and classification is to detect the event mentions
of target event types in plain text. Given an input text, an event detection
(ED) system should be able to identify whether the sentences contains events of
interest by means of the identification of event trigger terms (event identification)
and classify them into specific event types (event classification). For instance, in
the following sentence:
“Meridian National Corp. said it sold 750,000 shares of its common
stock to the McAlpine family interests, for $1 million, or $1.35 a share.”
According to the TimeBankPT corpus [
In this work, we employ TEFE to historical data, on the previously transcribed texts from the Monsoon Books, and evaluate its usefulness to the identification of historical junctures in historical corpora. 4
Each volume of Monsoon books encompasses more than 300 printed pages of narrative text. In this sense, applying a computational tool of event extraction enhances historical research to rapidly extract text information for a large collection of data. One way of studying this source is by organizing the events it describes. Language technology may help the reader with hints, extraction and quantification of these events. The system described in the last section was developed with the purpose of finding and classifying mentions to events, as well as identifying the participants of the events. The system receives an input sentence such as “Eu el-rey fac¸o saber aos que este alvaar´ virem que tenho entendido que pelo mau comcerto que tiveram as naus que, o anno passado de seiscentos e quinze, partiram do porto de Goa para este reino[...]” and identify that “partiram” (departed) is an instance of a Departure event (described by the Departing Frame) and that “as naus” (the ships), “porto de Goa” (Goa’s harbor) and “este reino” (this kingdom) are entities participating in such event, as depicted below: as naus que[...] |T h{ezme}
partiram trig|ger:Depar}ting {z do porto de Goa para este reino | So{uzrce } | G{ozal }
In this work we present an initial assessment of the application of this system in the source under study, to understand the needs for adapting the tool for language variants (from a diferent time span, in this case). Next we present some passages of the studied source, isolated in five examples, and what the system has produced as output, along with a discussion of the challenges to face.
Example 1: Eu el-rey fac¸o saber aos que este alvaar´ virem que tenho entendido que pelo mau comcerto que tiveram as naus que, o anno passado de seiscentos e quinze, partiram do porto de Goa para este reino, e por virem sobrecarregadas, em saindo da´quella barra comec¸aram logo a fazer agoa, o que foi causa de se perderem as naus Capitania e Sam Boaventura, e a Sam Philippe vir mui arriscada, e por esse respeito receber minha fazenda e meus vassalos noat´vel perda;
The passage mentions a communication made about bad mantainance of ships and the consequent loss of value related to overweighted ships coming from Goa to Portugal which sunk. In this passage, as presented in Table 1, 7 events were identified, related to the communication brought about bad maintenance of the ships, the leave of these ships from Goa, the start of the problems caused by the bad maintenance, which caused the loss of values due to the problems encountered with 3 of the ships.
Example 2: e porque convem muito a meu servci¸o saber-se com a considerac¸˜ao necessaria de como se procedeo no concerto e carga das ditas naus, e se ouve culpa de alguem de partirem tarde, hei por bem e mando ao Desembargador Gonc¸alo Pinto da Fonseca tire devassa na conformidade d´este alvara, sabendo mui particularmente a causa que houve pera as ditas naus virem sobrecarregadas e tam mal concertadas, e partirem tarde, e como se procedeo no concerto e carga d´ellas como se refere; In continuation from the previous passage, here there is a demand for the causes of the poor maintenance of the ships and how this was done, asking also for the identification or the responsible related to the delay of their departure, the causes for the overweight. The results of processing this passage on TEFE is depicted on Table 1. Two events identified were related to the requirement of information, and the third (removing) was in fact used as asking rather than removing, an error which might be explained due to verb ambiguity.
Example 3: e depois de tirada a dita devassa a mande serrada por vias nas primeiras naus que pera este reino partirem da´quellas partes, dirigida ao Conselho de minha fazenda, pera se ver n´elle e prover no caso como mais convier a meu servci¸o;
At this point, instructions are given on how the answer to the enquiry should be sent. The event “ver” (to see), depicted in Table 1, is identified but in this case the present stage of development of this tool is not flexible enough and comprises misunderstood meanings of the text, common in a Portuguese text with 400 years old.
Example 4: o que cumpriar´ sem duvida alguma, por este que valear´ como carta e n˜ao passaar´ pela chancellaria, o qual vai por tres vias. Francisco de abreu o fez em Lisboa a xbij (desasete) de fevereiro de seiscentos e desaseis. Diogo Soares o fez escrever.
In this particular example, the tool enables us to automatically understand the material authorship of the particular letter, identifying an agent responsible for the act of writing the document in a certain space and time, as can be seen in Table 1. In this analysis, it is not relevant to make an analysis of word statistics, since the documental corpora was not entirely digitally intervened yet. By the results shown in Table 1 we immediately perceive how the diverse documents present in the analysed volume of Monsoon Books were used to report to the metropolis or to India about the remote events relevant to the Portuguese administration of Estado da India and the Cape Route in the beginning of the seventeenth century. The most common event type is awareness underlining the role of communication that these documents performed. 5
This paper analyses the application of event identification, extraction and classification as a step for the study of historical sources. This initial study was required to know the current problems and help us shape further work. As such, we are now planning to annotate events in a portion of this historical source. Annotation will serve for adapting the tool for this specific temporal language variant and the required historic study needs, and also for more rigorous evaluation of the extraction provided by the tool. Regarding the language diferences, we plan to study whether normalization could improve the extraction of events. The adapted version of the tool will be used for further studies about this historical source. These tools may serve as reader´s guidance for the observation of historical sources contents. The tool enhances a more eficient process of information extraction of huge series of historical texts. From the examples shown, and regarding the nature of this source, we may consider uses such as helping researchers to find all events related to ships, actions of the Crown, conflicts, etc. If associated with other NLP tools it would be possible to create a list of the names of the ships, perhaps finding information about the travels made by them. It also serves the purpose to observe the geographical location of a person in a certain time and space and the type of actions they perform as historical characters. Even a statistical analysis of the type of events associated with time and places can trace new insights in the historical interpretation of a certain reality. We can detect patterns of human action by the most frequent event types on a certain time frame, as well as interpret the exceptional events and evaluate their historical meaning.If we are able to cross examine the interaction between the most frequent event types and the type of arguments they were most linked to in the future, we can easily perceive historical trends in the type of events which mostly concern the Portuguese administration in the Eastern part of the Portuguese empire.