A Network Analyss of Hermeneutc Documents Based on Bble Ctatons Hajme Mura (H_MURAI@valdes.ttech.ac.jp) Department of Value and Decson Scence, Tokyo Insttute of Technology -1-1, Ookayama, Meguro-ku, Tokyo 15-855 Japan Akfum Tokosum (AKT@ valdes.ttech.ac.jp) Department of Value and Decson Scence, Tokyo Insttute of Technology -1-1, Ookayama, Meguro-ku, Tokyo 15-855 Japan Abstract In order to handle systematc thought, we propose a new method of representng the man elements of a conceptualzaton n the form of a network based on the relatons between the ctatons for frequently-used propostons wthn a text corpus. Ths method makes t possble to automatcally extract the central components of a systematc thought and to analyze the relatonshps between these by usng the clusterng method. In the present study, we have constructed three networks of Chrstan dogma based on the wrtngs of St. Augustne, St. Thomas Aqunas and Pope John Paul II, and have analyzed the clusters to objectvely extract common elements and ndvdual characterstcs. Our network representaton and analyss method help to lay the foundatons for scentfc research n abstract felds of human thought, such as theology. Reflectng the ways n whch certan ndvduals perceved a partcular canoncal text, such as the Bble, these characterzatons of key conceptualzatons provde mportant nsghts nto the structure of the canoncal text tself. And. ths method can be appled to other feld where there are frequent repettons of propostons or sentences n order to objectvely analyze ther meanngs or nterpretatons. Keywords: knowledge representaton; ctaton analyss; Bble Background and Goals Background As an aspect of hgher cogntve functons, systematc thought has prmarly been nvestgated usng lteraturebased approaches, wth texts that are usually more abstract and subjectve n nature than scentfc papers. However, as systematc deas and thought nfluence all areas of human actvty and thnkng, the applcaton of NLP (Natural Language Processng) technques may provde a more objectve understandng of systematc thought. Recently, new methods are beng developed, such as the extracton of metrc characterstcs for text corpora (Lu, 1993; Whte & Mccan, 1989) specally usng co-ctaton analyss methods (Fnn, 004; Meyer, 004), text summarzaton through the automatc classfcaton of ctaton nformaton (Nanba & Okumura, 000). Wthn the feld of cogntve structure as well, new approaches are beng developed wth whch to symbolze and represent concepts n the form of networks such as a Conceptual Map (Rye, 00; Nelson & McKnney 1993). Methods that are capable of automatcally extractng the man elements of a conceptualzaton are partcularly mportant when handlng extremely large text corpus. By utlzng these new scentfc methods, we can (a) ensure the objectvty and replcaton of results, (b) handle large-scale data n a unform manner, and (c) reduce nformaton processng costs. We beleve that t s possble to analyze the abstract thoughts and value systems emboded wthn a text corpus wth such methods. In ths paper, we analyze a Chrstan text corpus. Tradtonal relgons have exerted great nfluences on humanty throughout hstory. Most relgons have at ther core some canoncal texts, wth the hermeneutcs, or nterpretatons, of the canon s also usually n text format. Thus, t s possble to represent key conceptualzatons wth the canoncal texts through ther objectve analyss. Goals Specfcally, the goals of ths study are to automatcally extract the man elements of a number of key conceptualzatons from a large-scale relgous text corpus and analyze the cluster constructon of them usng an objectve and replcable methodology. Ths, n turn, wll provde an objectve bass for the examnaton of systematc thought. Here we focus on the wrtngs of St. Augustne and St. Thomas Aqunas, two nfluental Church Fathers, as well as those of Pope John Paul II to extract essental teachngs of the Catholc dogma through hstorcal transton and to dentfy ndvdual characterstcs of hermeneutcs. Based on the patterns of the Bble ctatons wthn ther wrtngs, we created networks for frequently cted sectons of the Bble, and extracted the man elements and clusters of these, n order to compare a number of key conceptualzatons. The Canon and ts hermeneutc lterature The pllars of the Catholc value system are the Sacred Scrpture (the Bble), the Sacred Tradton, and the Teachng Offce (Paul VI, 1965; John Paul II, 199). Of these, Bble study s the soul of sacred theology. On the other hand, the Sacred Tradton and the Teachng Offce have mportant roles n nterpretng the Bble. Thus, Catholc dogma conssts of the Bble and ts nterpretatve lterature. By the complete and systematc analyss of the nterpretatons of each secton of the Bble and of the relatonshps between these, t s possble to 1565
vsualze the Catholc value system. The text data conssts of wrtngs by Pope John Paul II, posted on the Vatcan Web Ste (Englsh verson of Encyclcals, Motu propro, Apostlc Exohtaton, Homly, Apostlc Letter, and Speech etc. 315 fles), the wrtngs of St. Thomas Aqunas (9 ttles, ncludng Summa Theologca and Summa Contra Gentles ) and the wrtngs of St. Augustne (9 ttles, ncludng The cty of God and Confessons, and 1 homles). Methods Frst, the patterns of the Bble ctatons were compared for the complete corpus. Then, the relatonshps between the cted sectons of the Bble are extracted by co-ctaton analyss, from whch the ctaton networks were created. After that, the clusters of those networks are dentfed, and characterstcs of these networks are compared. Characterstcs of the Bble ctatons Comparsons of the three authors looked at ctaton dstrbutons, and at whch sectons and books of the Bble were ctatons most frequently taken from. Constructon of the ctaton networks based on coctaton analyss The Bble s separated nto unts of book, chapter and secton, and numbers are assgned to these unts. In ths study, a ctaton unt s defned as the smallest secton. Although there s some varaton n the lengths of these sectons, frequent-cted sectons are generally about 1 to sentences. Co-ctaton analyss was used to analyze these cted sectons. As Small clams, gven that co-cted texts have certan smlartes, the degree of smlarty between texts wll ncrease n proporton to the number of co-ctatons (Whte & McCan, 1989). Wthn the Catholc hermeneutc lterature, there are many Bble ctatons, makng t possble to measure the smlarty between two Bble sectons n terms of the frequency of ther co-ctaton. By connectng sectons that have hgh frequences of co-ctaton to form a ctaton network, t s possble to represent the complete doctrne. Although the ctng unt normally used n co-ctaton analyss s the whole document, because these texts are dvded nto smaller semantc unts accordng to Catholc tradton, ths unt was used as the present analyss. In lne wth one of the goals of ths study to capture the man elements of a conceptualzaton we have focused on the central parts of the networks, where the elements are both frequently cted sectons and for whch co-ctaton frequency s above a certan threshold (Mura & Tokosum, 004). 1: For the wrtngs of each author, the total number of ctatons ( v ) for a gven Bble secton, I, was calculated. Based on ths, t was possble to select V sectons, for whch v exceeded the ctaton threshold ( Th 1 ). The each couple of the V sectons were connected f the co-ctaton value ( e j ) exceeded a co-ctaton threshold ( Th ). v V v (1) Th 1 ej E v, v j V, ej Th () : Connectons were also made to sectons that ntervened between V sectons and also had co-ctaton values over the co-ctaton threshold ( Th ). v j V, vk, vl V, v V j, e jk, e jl E e jk Th, e jl Th, 3: Creatng the network by connectng V E V E (3) Extracton of the clusters wthn the ctaton networks Networks are usually composed of some clusters. We can understand the characterstcs of networks by extractng the clusters and analyzng these mutual relatonshps. There are many clusterng algorsms. In ths paper we extract clusters by shftng the co-ctaton threshold ( Th ). The frst mert s that the clusters reflect the strength of coctatons. Normal methods that extract some type of clques reflect only the network topology. The second mert s that we can change the sze of the clusters by shftng the coctaton threshold, makng t easer to extract clusters of approprate szes. Results Ctaton and Co-ctaton Patterns Table 1 presents all ctatons, all cted sectons, all coctatons, and all co-cted secton pars, together wth all sectons that nclude the Bble ctatons and the average ctatons per unt for the three authors. Whle the wrtngs of St. Thomas nclude more varous ctatons, those of John Paul II have lttle dversty n terms of the total ctatons. Table 1: Total ctatons. Augustne Thomas. John Paul II All ctatons 674 36015 3166 All cted sectons 8645 1181 8851 All co-ctatons 1584 800457 643708 All co-cted sectons 189353 75401 508118 All sectons 368 393 3444 Average ctatons 6.938 15.050 9.340 Table shows the most frequently cted books of the Bble. The numbers of each book are the ratos of the 1566
ctatons of that book dvded by total sectons of that book. Of these, 14 books are common to three authors (Bold). Ths result ndcates a smlarty n ther patterns of ctaton. The Gospel accordng to John s the most cted Gospel. And, John Paul II ctes more sectons from the Gospels than the others. However, analyss of the most frequently cted sectons, (Table 3), shows that common sectons are rather rare. There are only 5 common sectons (Bold: Jn1:14, Jn14:6, Ph:7, Ph:8, Rm5:5) among the top 40. Ctaton Networks and Clusters In comparng the three authors, the number of V elements and the sze of the maxmum connected component were adjusted by the thresholds. For St. Augustne, Th 1 =5, Th =4 and for St. Thomas Aqunas, Th 1 =7, Th =5 and for Pope John Paul II, Th 1 =41, Th =5. Fgures 1,, and 3 show the respectve maxmum connected partal graphs. The numbers wthn the [ ] are cted number, and the values V E V E are ndcated as follows: The clusterng threshold s selected so that the numbers of clusters n each network s smlar (n ths paper, clusters are 4 or 5), and thresholds were for St. Augustne 6, for St. Thomas 7, and for Pope John Paul II, 7. The results of clusterng are represented n Fgures 4, 5, and 6. Table : Top 0 most frequently cted books. Augustne Thomas John Paul II Rm 4.958 Rm 5.155 Ep 6.568 Ga 3.73 Ep 4.761 Jn 5.058 Ph 3.615 1Co 4.746 Ph 3.885 1Co 3.336 Ph 4.19 Lk 3.514 Ep 3.097 Ga 3.718 Mt 3.415 Mt.498 1Jn 3.667 Ga 3.396 1Jn.371 Jm 3.389 Col 3.368 Col.316 1Tm 3.336 1Jn 3.305 1Tm.83 1P 3.57 1Co 3.096 Jn. Jn 3.107 Rm.988 Co.08 Heb 3.066 1P.505 Tm 1.54 Col 3.053 Heb 1.749 Tt 1.478 Co.941 Tm 1.735 Ml 1.473 Mt.899 Ps 1.61 Jm 1.34 Tm.313 Mr 1.600 Gn 1.3 Tt.304 Ac 1.595 Ac 1.1 Ml.091 Rv 1.489 1P 1.10 P 1.934 1Tm 1.416 Th 1.149 Ws 1.68 Co 1.367 P 1.148 1Th 1.517 Tt 1.196 Table 3: Top 40 ctatons for all sectons of each books. Augustne Thomas John Paul II Mt6:1 84 Jn1:14 66 Mt8:0 13 Jn1:14 70 Rm5:5 55 Jn14:6 11 Rm5:5 68 Rm8:15 50 Jn17:1 111 Jn1:1 66 Ep3:17 48 Jn3:16 103 Rm5:1 64 Ga4:4 48 Mt8:19 9 1Co13:1 56 Heb11:6 47 Heb13:8 83 Jn1:3 55 Jn1:17 43 Lk4:18 8 Rm7:5 5 Rm1:0 43 Jn10:10 76 Ga5:6 51 Heb1:3 4 Mt5:40 74 1Tm:5 47 Ph:7 41 Jn1:14 69 Ga5:17 44 Ph:8 41 Gn1:6 68 Ph:6 43 Rm8:9 40 Rm5:5 66 Rm7:3 43 1Co13:1 39 Ga4:4 64 Rm7:4 40 Ep5: 39 Ph:8 63 Ph:7 39 Jn17:3 39 Mt5:14 63 1Jn1:8 38 1Co15:10 37 Gn1:7 59 Ws9:15 38 1Jn4:16 37 Jn13:1 59 Jn14:6 36 Jn1:18 37 Mt5:13 59 Mt5:8 34 1Jn3: 35 Lk5:4 58 Mt6:13 34 Is53:7 34 Ph:7 56 Rm7:18 34 1Co4:1 33 Jn0: 54 Jn1:9 3 Jn14:6 33 Ac1:8 54 1Co1:31 31 Ph3:1 33 Gn1:8 51 1Co4:7 31 Gn1:6 3 Lk1:38 50 1Jn3: 31 Rm6:3 3 Ep:14 49 Ac:4 31 Rm8:3 3 Mt5:35 49 Rm1:17 30 1Tm:4 31 Jn16:13 46 Mt5:41 9 Jn3:5 31 Jn0:1 46 Jn3:5 8 Rm5:1 31 Mr1:15 46 Rm10:3 8 Mt1:1 30 Mt5:36 45 Jn10:30 7 Rm10:10 30 1p3:15 45 Jn15:5 7 Ep:8 9 Jn:5 44 Rm1:0 7 Ex0:1 9 Ac1:14 43 Ph:8 6 Mt5:16 9 Ac:4 4 Rm7: 6 Mt11:9 8 Jn10:11 4 1Co3:7 5 Ws8:16 8 Jn19:7 41 Co4:16 5 Heb11:1 7 Ga4:6 41 Ac:3 5 Mt19:1 7 Jn15:5 41 Gn1:7 5 Rm6:4 7 Lk:3 41 Lk4:19 41 1567
Fgure 1: Ctaton Network for St. Augustne. Fgure : Ctaton Network for St. Thomas Aqunas. Fgure 3: Ctaton Network for Pope John Paul II. 1568
Fgure 4: The Ctaton Network for St. Augustne after Clusterng Analyss. Fgure 5: Clusterng Result of Ctaton Network for St. Thomas Aqunas. Fgure 6: Clusterng Result of Ctaton Network for Pope John Paul II. 1569
Dscussons Indvdual Characterstcs St. Augustne s network conssts of two relatvely separate parts; one ncludng Matthew 6:1-13 and Rome 7:18-5 and one coverng John 1:1-14 and Phlppans :6-8. Although hs two man themes are ncarnaton and the struggle wth sn, they are only weakly related. There are stronger connectons between the clusters wthn St. Thomas s network, where the central concepts are concerned wth ncarnaton (John 1:14, Phlppans :7-8, Ephesans 5:), the Trnty (Rome 5:5, 8:15, 9, Galatans 4:4), and love and fath (Ephesans 3:17, 1 John 4:16). The network for Pope John Paul II has a large cluster concerned wth spreadng the Gospel (Matthew 8:19-0, John 17:1, 0:0-1, Acts 1:8). Ths may reflect the presence wthn the corpus of hghly authortatve texts that are probably more representatve of Vatcan opnon. Common Characterstcs Common elements n the teachngs of the Catholc Church are a constant emphass on Jn1:14, Jn14:6, Ph:7, Ph:8, Rm5:5. Accordng to the clusterng results, Jn14:6, Ph:7 and Ph:8 compose common clusters n three author s networks. These sectons are about Incarnaton. Jn14:6 s I am the way and the truth and the lfe. No one comes to the Father except through me. And Rm5:5 s hope does not dsappont, because the love of God has been poured out nto our hearts through the holy Sprt that has been gven to us. These sectons are ncluded n dfferent clusters on each network. Pope John Paul II ncludes both wthn a cluster about Incarnaton, but St. Augustne ncludes Rm5:5 wthn a cluster that focuses on sn. So, we may regard these two nterpretatons as beng dfferent. From the ctaton analyss and ctaton networks, we may conclude that the most mportant teachng of the Catholc Church common throughout ts hstory s the Incarnaton of Chrst. Conclusons By representng the frequently-used elements of the network structure, t s possble to objectvely perceve the complete systematc framework of ths complex theology. Ths s functonally smlar to automatc summarzaton for meanng. Moreover, the network structure s also useful for extractng and numercally analyzng semantc dfferences n felds where complex nterpretatons are requred. Ths paper demonstrates that our new method s effectve by comparng abstract thoughts and extractng common and unque elements n those abstract thoughts objectvely. Our network representaton and analyss provdes a scentfc bass for research n more abstract felds, such as theology. To the extent that the extracted structures are reflectng the semantc structures perceved by the nterpreter, t also provdes a means of capturng the nner structure of the Bble. Accordngly, ths network constructon and clusterng method can be appled to other feld where there are frequent repettons of propostons or sentences. References Douglas L. Nelson & Vanesa M. McKnney, (1993) Implct Memory Effect of Network Sze And Interconnectvty On Cued Recall, Journal of Expermental Psychology: Learnng, Memory, and Cognton, 747-764. Fnn R. Forsund & Nkas Sarafoglou, (004) The tale of two research communtes: The dffuson of research on productve effcency, Internatonal Journal of Producton Economcs, In Press, Corrected Proof, Avalable onlne. Hajme Mura & Akfum Tokosum, (004) A Network Representaton of Hermeneutcs Based on Co-ctaton Analyss, WSEAS Transactons on Informaton Scence and Applcatons, l1, 6, 1513-1517. Hdetsugu Nanba & Norko Kando & Manabu Okumura, (000) Classfcaton of Research Papers usng Ctaton Lnks and Ctaton Types: Towards Automatc Revew Artcle Generaton, The Amercan Socety for Informaton Scence (ASIS) the 11th SIG Classfcaton Research Workshop, Classfcaton for User Support and Learnng, 117-134. James Rye & Peter Rubba, (00) Scorng Concepts Maps: An Expert Map-Based Scheme Weghted for Relatonshps, School Scence and Mathematcs, 33-44. Lu, M, (1993) Progress In Documentaton The Complextes of Ctaton Practce: A Revew of Ctaton Studes, Journal of Documentaton, 49, 4, 370-409. Martn Meyer & Tago Santos Perera, (004) Olle Persson and Ove Granstrand, The scentometrc world of Keth Pavtt: A trbute to hs contrbutons to research polcy and patent analyss, Research Polcy, 33, 9, 1405-1417. Pope John Paul II, (199) Catechsm of the Catholc Church, http://www.vatcan.va/archve/eng0015/_index.htm. Pope Paul VI, (1965) Documents of the II Vatcan Councl De Verbum, http://www.vatcan.va/archve/hst_ councls/_vatcan_councl/documents/vat-_const_ 19651118_de-verbum_en.html. Whte, H. D. & Mccan, K. W., (1989) Bblometrcs, Annual Revew of Informaton Scence and Technology, 4, 119-186. 1570