Discovering Land Transaction Relations from Land Deeds of Taiwan

  1. 1. Shih-Pei Chen

    CSIE Department - National Taiwan University

  2. 2. Yu-Ming Huang

    CSIE Department - National Taiwan University

  3. 3. Hou-Ieong Ho

    CSIE Department - National Taiwan University

  4. 4. Ping-Yen Chen

    CSIE Department - National Taiwan University

  5. 5. Jieh Hsiang

    CSIE Department - National Taiwan University, Research Center for Digital Humanities - National Taiwan University

Work text
This plain text was ingested for the purpose of full-text search, not to preserve original formatting or readability. For the most complete copy, refer to the original conference program.

Discovering Land Transaction Relations from Land Deeds of Taiwan
Chen, Shih-Pei, Department of Computer Science, National Taiwan University,
Huang, Yu-Ming, Department of Computer Science, National Taiwan University,
Ho, Hou-Ieong, Department of Computer Science, National Taiwan University,
Chen, Ping-Yen, Department of Computer Science, National Taiwan University,
Hsiang, Jieh, Department of Computer Science, National Taiwan University; Research Center for Digital Humanities, National Taiwan University,
Land deeds were the only proof of ownership in pre-1900 Taiwan. They are indispensable for the studies of Taiwan’s social, anthropological, and economic evolution. We have built a full-text digital library that contains more than 30,000 land deeds. The deeds in our collection range over 250 years and are collected from over 100 sources. The unprecedented volume and diversity of the sources provide an exciting source of primary documents for historians. But they also pose an interesting challenge: how to tell if two land deeds are related.

In this paper we describe an approach to discover one of the most important relations: successive transactions involving the same property. Our method enabled us to construct over 3,300 such transaction pairs. We also introduce a notion of land transitivity graph to capture the transitivity embedded in these transactions. We discovered 2,219 such graphs, the largest of which includes 103 deeds. Some of these graphs involve land behavior that had never been studied before.

Until the turn of the 20th century, hand-written land deeds were the only proof of transaction of lands in Taiwan. Such a deed may involve activities such as selling/buying, lending of land to smaller farmers, dividing the land among children or shareholders, and cultivation permits. The deeds usually follow, depending on their types, a typical but not standard format, and are drown up in ad hoc manner. Indeed, even the name of the location may be written in a local convention unfamiliar to the outsiders.

While each land deed may have significance only to its owner, a large collection of them provides a fascinating glimpse into the pre-modern Taiwanese grassroots society. Historians have studied them to investigate the economic activities, community development, and the relationship among the various ethnic groups (Chen, 1997; Ka, 2001; Shih, 2001; Hong, 2005).

In the past few years we have built a full-text digital library of primary historical documents of Taiwan called THDL (Taiwan History Digital Library). Among its corpuses is a collection of over 30,000 land deeds, spanning from 1666 to the first decade of the 20th century, and collected from over 100 sources of origin (Hsiang, Chen, Tu, 2009). This collection is unprecedented in terms of volume, time span, geographic distribution, and variety. While THDL presents an exciting source of primary materials for historians, it also poses a challenge: how to find the relationship between two land deeds, or, how to find all the land deeds involving the same piece of land. Although it was customary to hand down earlier deeds to the new owner during the transaction of land, most of these links were broken when the Japanese, during their colonial rule of Taiwan between 1895 and 1945, modernized the land management system (Li, 2004). That is because the officials only recorded the last deed as the proof of ownership but ignored the previous ones. Consequently many of the older deeds were either destroyed or (later) sold as collector’s items because they had lost their original value.

In this paper we present a semi-automated method to discover the transaction relations among land deeds. We shall focus on two important relations: successive transaction pairs and allotment agreements. We further connect the transitive activities on the same piece of land into a concept called land transitivity graph, which captures the history of the land over time. The largest such graph that we found has led to a discovery of a new type of land use that had never been observed before.

Discovering Land Transaction Relations
We start by describing the two relations among land deeds that our method tries to capture.

Successive transaction pairs: A piece of land could be sold from A to B, then from B to C. In this case there should be two land deeds recording the two transactions. We call them a successive transaction pair. Note that the situation could be rather complicated. For instance it could have been B’s son who sold it to C. If B divided the land among his descendents, the first selling transaction and the ensuing allotment agreement (see below) also form a successive transaction pair.

Allotment agreements: An allotment agreement is a deed that records how a land is divided among the owner’s descendants or among the shareholders. In both cases the usual practice is to first divide the land into several parts, then to have each participant drawing from the lot. Once the decision is agreed upon, an agreement is written, and several copies are made and given to each person involved. In the case of division among shareholders, the allotment agreements should be preceded by a cultivation permit, a permission from the government to allow a group of people to cultivate the land. In this case, the cultivation permit and the ensuing allotment agreement also form a successive transaction pair.

Fig. 1 The process for discovering land transaction relations

Full Size Image

To tackle this problem of finding successive transaction pairs, we developed a 3-step semi-automatic process (Fig. 1). We first used text processing technology to extract features of each land deed from its metadata and full text. Such features include the transaction type, the general location of the land and the four reaches (boundaries identifying the land via some obscure way such as “bordering Lee’s house on the south,” “a large camphor tree on the west,” etc), the names of the people involved in the transaction and their roles (seller, buyer, scrivener), description of the source of the land (how and when the current owner obtained it), the size, the price, and the amount of taxes paid (Lu, 2008; Huang, 2009). Fig. 2 is an example of a typical land deed. We designed an XML format to hold this information (Fig. 3). Second, we defined rules to identify deeds that may be related. Fig. 4 shows the rules we used for identifying the successive transaction pairs. We then wrote a program to compare every pair of land deeds in THDL to see if any pair satisfied the rules. Finally, we give all the pairs produced to human expert to verify.

Fig. 2 An example of a typical land deed of Taiwan

Full Size Image

Fig. 3 The features of a “selling” type of land deed, stored in XML

Full Size Image

We further remark that a criterion that allows certain degree of fuzziness was used when performing matching. This is because the names used in different deeds may sometimes be slightly different even if they are the same place or person (Huang, 2009).

Fig. 4 The rules for identifying successive transaction pairs

Full Size Image

The precision rate of the algorithm for successive transaction pairs is 63.9% and that for allotment agreements is 94.4%. We have found 2,409 successive transaction pairs and 878 sets of allotment agreements among the 30,820 land deeds in THDL (Table 1).

Table 1 The result of reconstructing land transaction relations among the land deeds in THDL

Full Size Image

Among the former, 358 are cross-generation (A sold to B, and B’s descendent sold to C). Some of the pairs/sets are from different sources (the “cross sources” column in Table 1), and are quite impossible to find manually. Some others are from the same source but are not adjacent to each other in their original order. These are also difficult to identify by hand.
Land Transitivity Graphs
When further examining the transaction pairs, an interesting transitive phenomenon emerged. There may be a deed of A selling a piece of land to B, and some years later B divided the land among his sons, then one of them, C, rented it to D to farm. Such transitive activities on the same piece of land could last for decades. By connecting all these transactions into a graph, it may capture the evolution of a property over time.

This is exactly what we did. We call these graphs land transitivity graphs. Using the relations we discovered early, we came up with 2,219 such graphs. The result is listed in Table 2.

Table 2 The land transitivity graphs constructed among the land deeds in THDL

Full Size Image

Fig. 5, the third largest graph, contains 36 deeds, dating from 1850 to 1910. The head of the family, Liao Jiafu (廖佳福), was among the shareholders who received a cultivation permit from the Qing government, and obtained this piece of land through allotment in 1850 (the first deed).

Fig. 5 The 3rd large graph, containing 36 deeds

Full Size Image

Liao farmed the land for 50 years and divided it among his descendents in 1901 (the second deed). The rest of the deeds described the various activities such as further divisions or selling in the next 10 years. By 1906, only 2 of the 8 divided pieces of land remained in the Liao family.
Fig. 6 The largest graph, containing 103 deeds

Full Size Image

Fig. 6 is the largest land transitivity graph with 103 deeds. Tu, a historian, studied this graph and discovered that the deeds involved demonstrated a unique case of land use that had never been studied before (Tu, 2010). It is unlikely for human to notice this possibility without the computer-generated transitivity graph.

To help historians take advantage of these graphs, we developed an integrated environment to analyze the information embedded in each graph (Fig. 7). In addition to the graph itself and its zoomable navigation facility, we also added tag cloud, chronological distribution, and a location map.

Fig. 7 The integration environment for land transitivity graphs

Full Size Image

Concluding Remarks
Land deed research has been an important topic among historians of pre-1900 Taiwan. In this paper, we presented a method to discover the transaction relations among the 30,820 land deeds in THDL, the largest existing full-text database of land deeds. Our method discovered 2,049 successive transaction pairs and 878 sets of allotment agreements. They, in turn, are transformed into 2,219 land transitivity graphs, each of which describes the transaction evolution of a piece of land. One such graph has already led to the discovery of a unique pattern of land development that had not been studied before (Tu, 2010). We feel that our work demonstrates how IT tools can be used to help historians conduct research that could not be done otherwise.

Chen, C. K. 1997 Taiwan’s aboriginal proprietary rights in the Ch’ing period: Bureaucracy, Han tenants and the transformation of property rights of the Anli Tribe, 1700-1895, Academia Sinica Taipei

Hong, L.W. 2005 A study of aboriginal contractual behavior and the relationship between aborigines and Han immigrants in west-central Taiwan, 1 Taichung County Cultural Center

Hsiang, J., Chen, S. P., Tu, H. C. 2009 “On building a full-text digital library of land deeds of Taiwan, ” Digital Humanities 2009 Conference, Maryland, June 22-25, 2009 85-90

Huang, Y. M. 2009 On reconstructing relationships among Taiwanese land deeds. Master thesis, National Taiwan University Taipei, Taiwan

Ka, C. M. 2001 The aborigine landlord: Ethnic politics and aborigine land rights in Qing Taiwan, Academia Sinica Taipei

Li, W. L. 2004 “Land deeds and land administrative documents—Interpreting the Archives of the Japanese Taiwan Governor-Generals, ” Taiwanese History Research, 11 (2) 221-240

Lu, C. C. 2008 Automated Classification of Taiwanese Land Deeds. Master thesis, Taiwan University Taipei, Taiwan

Shih, T. F. 2001 Local society in Qing Taiwan, Cultural Affairs Bureau Hsinchu

Tu, F. E. 2010 “Environmental Change, Land development and Dispute over Property Rights in southern Taiwan (1890-1920), ” The Sixth Conference of Taiwan Colonial Government Archives, Taiwan Historica,

If this content appears in violation of your intellectual property rights, or you see errors or omissions, please reach out to Scott B. Weingart to discuss removing or amending the materials.

Conference Info


ADHO - 2011
"Big Tent Digital Humanities"

Hosted at Stanford University

Stanford, California, United States

June 19, 2011 - June 22, 2011

151 works by 361 authors indexed

XML available from (still needs to be added)

Conference website:

Series: ADHO (6)

Organizers: ADHO

  • Keywords: None
  • Language: English
  • Topics: None