Mining and Visualizing the Narration Tree of Hadiths

493

Chapter 29

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions) Aqil Azmi King Saud University, Saudi Arabia Nawaf Al Badia General Organization for Social Insurance, Saudi Arabia

Abstract Hadiths are narrations originating from the words and deeds of Prophet Muhammad. Each hadith starts with a list of narrators involved in transmitting it. A hadith scholar judges a hadith based on the narration chain along with the individual narrators in the chain. In this chapter, we report on a method that automatically extracts the transmission chains from the hadith text and graphically displays it. Computationally, this is a challenging problem. Foremost each hadith has its own peculiar way of listing narrators; and the text of hadith is in Arabic, a language rich in morphology. Our proposed solution involves parsing and annotating the hadith text and recognizing the narrators’ names. We use shallow parsing along with a domain specific grammar to parse the hadith content. Experiments on sample hadiths show our approach to have a very good success rate.

INTRODUCTION Hadiths are oral traditions relating to the words and deeds of Prophet Muhammad. The traditional Muslim schools of jurisprudence regard hadith as an important tool for understanding the Qur’an DOI: 10.4018/978-1-60960-741-8.ch029

and in all matters relating to jurisprudence. The hadith consists of two parts: the actual text of the narrative, known as matn (‫ ;)المتن‬and the chain of narrators through whom the narration has been transmitted, traditionally known as isnad (‫)إسناد‬. The isnad consists of a chronological list of the narrators, each mentioning the one from whom they heard the hadith all the way to the prime

Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

narrator of the matn followed by the matn itself. The isnad system began during the lifetime of the Prophet and was used by the companions in transmitting the hadith. The political upheaval around 655 CE/35 AH1 gave birth to the forgery of traditions in the political sphere, in order to credit or discredit certain parties. So, scholars became more cautious and began to scrutinize, criticize and search for the sources of information and that gave boost to the importance of isnad (Azami, 1978, pp. 246-7). And this gave birth to a new science, ‘Ilm al-Jarh wa al-Ta‘dil. In the minds of hadith scholars there are several factors that contribute to the overall grading of a hadith: the individual narrators involved, the transmission chain itself, and the supporting statement from all available evidence. Typically a hadith scholar will end up consulting many volumes on narrator’s biographic information for grading a single hadith. These books classify the narrators on their morality and their literary accuracy. Next, the chain of the transmission must not be broken, as a broken chain means a major defect in isnad. How does the hadith scholar decide this? By ensuring that there was an ample overlapping time between each pair of narrators in a chain to have met during their lifetime. Again this information is dug from narrator’s biography. For more detail on the subject the reader is referred to the work of Azami (1977). In this paper we report on a software tool that will automatically generate the transmission chains of a given hadith, graphically rendering its complete isnad tree. Such a tool is useful for the students of hadith to study how a certain hadith has been propagated, while a hadith scholar will find it valuable for his work on grading the hadith. We tested our system on many sample hadiths and the outputs were verified by hadith scholars. Overall a success rate of slightly over 85% was achieved.

494

Background We will start by looking at what a hadith is. In the subsequent discussion we will be quoting Arabic text. For convenience it will be followed by transliteration and English translation as well. There are several Arabic text transliteration schemes; we for one will be using the Buckwalter Arabic transliteration (“Arabic Transliteration,” 2002). Even though the Buckwalter transliteration is not intuitive and lacks readability, it has been used in many publications in natural language processing and in resources developed at the Linguistic Data Consortium (Habash, Soudi, & Buckwalter, 2007). The main advantages of the Buckwalter transliteration are that it is a strict one-to-one transliteration and that it is written in ASCII characters. The English translation will be based on (Al-Qushairy & Siddiqi, 1972). Before proceeding further, we feel it is necessary to write a few lines about Arabic for those who are not familiar with the language. The Arabic used in hadith is known as Classical Arabic, the Arabic of the Qur’an and early Islamic literature (7th – 9th century CE). However, this classical Arabic can be easily read and understood by anyone familiar with Modern Standard Arabic (“Modern Standard Arabic,” n.d.).

In Depth Look at Hadith Hadiths range in size from a few lines to a few hundred lines with the majority being five to six lines long. As an example of a hadith (the original Arabic followed by Buckwalter transliteration and English translation), see Figure 1. This is a hadith with a single chain of narrators, however, not all hadiths have such a simple chain as we will later see. The hadith corpus is quite huge. Early on, hadith scholars compiled it into six major collections; the bracketed numbers following the name of the collection refer to the number of hadiths in the compilation (from www.islamweb.net): Sahih

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Figure 1.

of Bukhari (7397); Sahih of Muslim (12000); Sunan of Abu Daud (5274); Sunan of Tirmizi (3956); Sunan of Nasa’i (5758) and Sunan of Ibn Majah (4341). There are other lesser known large collections (e.g. Muwatta’ of Imam Malik and Musnad of Ibn Hanbal).

does offer a limited option to render the narration tree of a hadith. Nonetheless, these graphs have been manually pre-compiled and are hardwired into the hadith database. Our scheme will render the narration tree on the fly offering greater flexibility for future expansion.

Significance of this Work

Problems and Challenges

There is a huge literature that deals with hadith and hadith related subjects (e.g. narrators’ biographies). And that literature is still growing. Many hadith scholars devoted their entire lives serving this literature. Their efforts have been purely manual. Of late scholars started to realize the importance of computers in this field. So now many of the hadith compilations exist in computer readable format, mainly as plain text, web contents or in proprietary locked databases. The only possible way to search these contents digitally is through the primitive search capability these tools provide or through traditional search engines, obviously both of which are incapable of analyzing and understanding the isnad context. Then, there exist some commercial products such as Hadith Encyclopedia by Sakhr. A software that was meant for the layman and hadith students. It

Processing natural languages has long been a hot topic of research. Regardless, today’s programs are still not able to parse sentences and understand adequately the precise context as humans usually do. Basically, this is because natural languages tend to have substantial ambiguity in the way they are used, and consequently are very hard to process. Arabic, a morphologically rich language, poses some additional challenges. Here we list some of the peculiarities in the hadith text that pose a challenge. 1. Anecdotal hadith verses deeds hadith: The hadiths relating to the words of the prophet tend to have a simple structure: name → name →…→ name → Prophet Muhammad (end of isnad) said: some saying of the prophet (end of matn). An example is shown in Figure 2.

495

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Figure 2.

On the other hand, hadiths involving the Prophet’s deeds or customs are more involved (see Figure 3). Here both Bilal and Ibn Umm-Maktoom are not part of the isnad even though they are both companions of the prophet. 2. A network of narration chains: When compiling hadith collections, some of the hadith scholars combined multiple isnad chains into a single hadith text provided they all shared the same matn, often marking the beginning of a new isnad chain by the Arabic letter (‫)ح‬, for example, see Figure 4. Figure 3.

496

For brevity, the isnad part of the hadith in Figure 4 has been trimmed. The partial narration tree is shown in Figure 5. Anas, (narrator no. 5, 10 and 15) is the prime narrator (i.e. the one who reported this hadith directly from Prophet Muhammad). 3. Identifying hadiths parts (isnad and matn): Sometimes finding the border point of the isnad and matn is not a trivial task. There are hadith cases where the boundary is difficult to determine. 4. Resolving ambiguity in the narrator’s names: In the hadith literature it is common

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Figure 4.

to refer to the same narrator with different names or alias. For example, see Figure 6.

X who heard his uncle” without mentioning the latter’s name (see Figure 7).

Three different names for the same person. Sometimes we find that referencing terms (e.g. his father, his mother, his grandfather...) are used in the narration context, for example: “narrator

5. Hadith referring to another: Another common practice in the hadith literature is to list the isnad and perhaps a partial matn while referring to the matn of an earlier hadith. For example, see Figure 8.

Figure 5. Partial isnad graph. The numbers refer to individual narrators in the hadith. Note that nodes no. 5, 10 and 15 all refer to the same prime narrator.

The matn of this hadith refers to the matn of an earlier hadith (most likely with a different narration tree) that has Yunus and Malik as narrators.

Related Work Though there are some commercial products that have computerized some aspects of hadith, nonetheless, as we noted earlier, these have been manually pre-compiled and locked into proprietary databases. And as such few research papers have been published in the area. In Al Asaneed Tree (Al-Osaimi et al., 2008) the authors worked on an algorithm that draws

497

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Figure 6.

Figure 7.

Figure 8.

the narration tree. Their system requires the user to enter individual narration chains which it will later draw into a tree form. It differs from our system where we semantically parse the hadith contents and draw the corresponding isnad tree. An interesting and somewhat related work (AlSalman, Al-Ohali, & Al-Rabiah, 2006) titled, “A Semantic Parse and Meaning Analyzer,” attempts to reveal the word sense ambiguity by building a semantic parser supported by a statistical semantic analyzer. The authors of this paper have exerted

498

great effort to build Arabic language grammar as rules and constraints represented in Backus-Naur Form (BNF) notation (Naur, 1960). This paper has inspired us to try to build a grammar for hadith and perform parsing to create a hadith parse tree.

Proposed System Our proposed algorithm is shown in Figure 9. It consists of a sequence of finite-state based pro-

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

cesses, which constitute the procedure to analyze the hadith narration chain(s). Our focus will be on natural language processing techniques to come up with a domain specific grammar for hadith. Next we go through each phase in more detail.

Preprocessing Phase We need to lexically analyze the hadith text. This process is considered a preprocessing phase for the parser. Prior to passing the hadith to the lexical analyzer, we remove some of the unnecessary characters (parser noise characters). The stages of preprocessing phase are shown in Figure 10. Sample Arabic input (text with full diacritical markings) and the output of the preprocessing phase is shown in Figure 11. Note that we removed the diacritical marking since they are not used in this version of the system.

Grouping and Tagging Sets of Words into Categories Shallow parsing is a natural language processing technique that attempts to provide some sort of machine understanding of the structure of a sentence but without parsing it fully into a parse tree form. Shallow parsing is also referred to as the task of chunking. The shallow parser produces and categorizes sentences into a series of words that have something in common. For instance, categorizing the sentences into noun, verb, preposition phrase, adjective, clause and other phrases (Stav, 2006). There is a noticeable difference between shallow parsing and full parsing in that the output of shallow parsing consists of a set of words that do not overlap and do not contain other chunks (non-recursive). We use shallow parsing to resolve two of the earlier mentioned challenges (see § Problems and challenges). For the deeds hadith it is used to remove unwanted sentences from the hadith that are not important to the full parsing process; it is also used to determine the border point of the isnad and matn.

Figure 9. Algorithm to analyze and visualize hadith narration tree

Figure 10. Stages of preprocessing phase

499

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Figure 11.

There are few technical research studies on the Arabic language. Furthermore, Arabic is known for its variation and richness which makes building linguistic tools a challenging task. Thus, we will only focus on our research scope. We are interested in NP (Noun Phrases) chunking and more specifically, narrator’s names. Therefore we will build a set of annotations to denote the chunks/categories that we intend to use as input for the full parser (hadith grammar). We devise the following notations to tag sentences: MT (sentence represents the matn); PM (sentence represents Prophet Muhammad); NN (sentence represents the Narrator’s name); and CC (sentence represents closed class words). The closed class is the set of words used as transmission terms in hadith literature, e.g. ‫( حدثنا‬haddathana), ‫أخبرنا‬ (akhbarana), ‫( قال‬qaal) …etc. Let us take an example to better illustrate the main concept of the shallow parser. Consider the hadith in Figure 12 and the corresponding output as produced by the shallow parser. As demonstrated by the previous example shallow parsing can deal with contents that contain noise characters very well. Noise is often left

500

outside chunks, and does no further damage; while in full parsing the parser will attempt to use it in deciding on higher-level parsed tree nodes. Next, we will see how to implement a shallow based parser. For the shallow parser we used a MemoryBased Learning model (Stanfill & Waltz, 1986). Memory-Based Learning (MBL) is a form of supervised, inductive learning from examples. The examples represent a vector of features assigned to a certain category. While training, the set of examples is streamed to the classifier and then added to the memory (knowledge base). During the test, a set of untrained data (new data) is presented to the classifier. For each test case, the distance is computed with all cases available in the memory. The nearest case category is used to predict the class of the test case. There are many algorithms in AI that touch some aspects of MBL, such as example-base (Peirsman, 2006), similarity-based (Zavrel & Daelemans, 1997), lazy based (Daelemans, van den Bosch, & Weijters, 1997) … etc. We will be concentrating on the similarity based approach in this paper. The performance of MBL depends on the similarity metric (distance metric). According

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Figure 12.

to (Veenstra & Buchholz, 1998) the most basic metric for patterns with symbolic features is the overlap metric given by:

where C is the set of class labels, P(𝜐) is the probability features vector, and Vf is the set of values for feature f The entropy of the class labels, H(C) is given by:

n

∆(X ,Y ) = ∑ d(x i , yi )

H (C ) = −∑ P (c) log P (c)

i =1

o ∈C

where ∆(X,Y) is the distance between patterns X and Y represented by n features, while δ is the distance per feature: δ(xi,yi) = 0 if xi = yi and 1 otherwise. The above metric simply counts the number of mis/matching feature values in both patterns. The classifier will consider the features equally important. The features are weighted using Information Gain (IG), which measures how much information it contributes to our knowledge of the correct class label. The Information Gain of feature f is measured by computing the differences in uncertainty between the situation without and with knowledge of the value of that feature, and is given by the following equation (Daelemans, van den Bosch, & Weijters, 1997): wf =

H (C ) − ∑ u∈V P (u) ⋅ H (C | u) f

si( f )



The split info si(f) is included to avoid a bias in favor of features with more values. It is defined as: si( f ) = −∑ P (u) log P (u) u ∈Vf

ib1-ig (Daelemans & van den Bosch, 1992) is

a memory based learning algorithm that is used to build a data base of cases during the learning. The case is represented as a vector of features. The classification works by matching the test case (new) to all the test cases (training set) and calculating the distance between the new case and the case in the memory. In ib1-ig is a weighted sum of the distances per feature. The distance will be zero when the values of both cases for this feature are equal and one otherwise. The ib1ig algorithm searches for the nearest neighbors which is an expensive operation. So to achieve 501

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

a fast chunking we have selected the Information Gain Tree (IGTree) model. According to the study in (Veenstra & Buchholz, 1998), IGTree gives a high measure of precision and recall as well as good accuracy. The IGTree combines two algorithms; one for constructing decision trees, and one for retrieving classification information from these trees.

Hadith Grammar The grammar is domain specific grammar in that it is only meant to parse hadith contents. This grammar requires a preprocessing phase which we refer to as “Lexical analysis” (Silberztein, 1997). The processing will remove the unneces-

Figure 13.

502

sary characters such as punctuations, extra white spaces, newlines … etc. As we stated earlier, full parsing is very sensitive to noise (typographical errors, repeated phrases, corrections etc). In comparison, chunking deals with noise very well; noise is often left outside chunks, and does no further damage, whereas in full parsing the parser will attempt to use it in deciding on higher-level parsed tree nodes. In pursuing our effort to fully parse hadith contents we have created this domain specific grammar (Figure 13), setting rules and constraints to help prevent part of the context ambiguity. The grammar rules and constraints are described in extended Backus-Naur Form (EBNF) (“Extended Backus-Naur Form,” n.d.). The EBNF is a

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

meta-syntax notation used to express context-free grammars (CFG) (Chomsky, 1956). It is an extension to the regular Backus-Naur Form (BNF). In writing the grammar in EBNF we follow the ISO standard (International Standards Organization [ISO], 1996), in addition we use ⊔ to denote the white space.

Testing the Hadith Grammar For testing purpose, we would like to apply our domain specific grammar on sample hadiths by

drawing the derivation tree. For that we will rewrite the grammar in CFG notation which will simplify the task of drawing the derivation tree. This time however, the grammar will be written in Arabic. In fact the original hadith grammar was also written in EBNF form in Arabic. With hadith text in Arabic, the Arabic CFG will naturally flow with the text from right to left making it easier to trace (see Figure 14). Figure 15 shows a successful derivation tree for the isnad and matn when the grammar is ap-

Figure 14. Hadith grammar in CFG notation

503

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Figure 15. A successful derivation tree. Only partial tree shown (full tree too big to fit)

Figure 16.

504

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

plied on the following preprocessed hadith shown in Figure 16. Only part of the derivation tree is shown as the full derivation tree is too big to fit into a page. Unfortunately, there were some cases where the text contains something we call “parser noise words.” These are words that are not accurately parsed using the grammar. Figure 17 is one such example. I have grayed out the entire text leaving the offending part in black. According to the hadith grammar, this portion of the hadith occurs in the Isnad part and so it is processed by it. But if we look closely, we see this portion of the hadith does not comply with the grammar rules and constraints set by the Isnad

part of the grammar. Detecting such noises in the preprocessing is not an easy task. In fact, there are some hadith cases where the context is very ambiguous even for the specialist readers. One possible solution is to deal with this kind of problems in post-processing phases.

Graphical Rendering of the Isnad Tree One main objective of this work is to graphically represent the hadith narration flow diagram, a directed graph where nodes represent individual narrators. Going through different hadiths we noticed cases where a single narrator narrates to

Figure 17.

Figure 18.

505

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

several others in the chain (one-to-many). As an example, note the term ‫( كلهم‬meaning all) in the example of Figure 18. Consequently there are also cases where a narrator hears the same hadith from two or more narrators in the hadith narration chain (many-toone). So the result is too many narrators with complicated intersections that resemble a general tree. For the isnad to be readable on the display, the nodes need to be laid out perfectly with the minimum number of intersections and links crossing over. Although there are several techniques in graph theory that address the obstacles surrounding nodes representation, there is still a need to come up with new algorithms to better enhance the readability and usability of the represented graph. Figure 19 (the full version of the isnad chain shown in Figure 5) illustrates a non trivial narration tree of a hadith. We will leave the detail of the isnad tree graphical display algorithm as it is outside the scope of this chapter. One thing we did not address in the current work is resolving the ambiguity in narrators’ names (see § Problems and challenges). A list of equivalent names (different names or alias representing the same narrator) at best partly solves the problem. For a comprehensive

solution we do need a huge database of studentteacher pairs.

evaluation and results For testing purposes we picked 32 hadiths from the simple cases and 55 from the hard cases, a total of 87 different hadiths. The hadiths were picked from Sahih of Bukhari and Sahih of Muslim since they are considered the most reliable of the collections. By simple cases we mean those hadiths that has a single narration chain, whereas the hard cases have more than one narration chain. The authors would like to thank Prof. Emeritus M.M. Al-Azami, a hadith scholar, for helping us verify the generated trees. The program successfully generated the narration chain of 28 hadiths from the simple cases and 48 hadiths from the hard cases. An overall success rate of 87%. Figure 20 is a screen shot of the narration tree of a hard case hadith. Figures 21 and 22 show the narration trees of two hard case hadiths which were successfully generated and rendered. In fact Figure 21 is the full narration tree for the hadith whose partial isnad graph was featured in Figure 5.

Figure 19. A non trivial narration tree. The leftmost node is the prime narrator of the hadith.

506

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

An example of a simple case hadith which our program failed to generate a correct narration chain was discussed as an example for inaccurate parsing (see § Testing the hadith grammar).

Conclusion and future research directions Working on Arabic text processing is an abstruse topic, because of the Arabic language remarkable richness in derivations, vocabularies and grammatical structures. In this paper we tackled one

Figure 20. Screen shot of a successfully generated narration tree of the hadith (in upper left window). Prime narrator is the leftmost node.

Figure 21. Successfully generated and rendered narration tree of a hard case hadith. This is the full isnad chain of the hadith featured in Figure 5. The prime narrator is the leftmost node in the figure.

507

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Figure 22. Another example of a successfully generated and rendered narration tree of a hard case hadith. The leftmost node is the prime narrator.

of the most celebrated classical Arabic literatures, the sayings and deeds of Prophet Muhammad. Our focus was on automatic generation and graphical visualization of the narration tree of the hadith. This process involved creating natural language lexer, perform shallow parsing, parsing, build syntactic analyzer and finally graph presenter that displays the narrators’ chain graphically. We believe our contribution will definitely help researchers in this field especially the grammar to parse the isnad. Our domain specific grammar is built from the ground up and was derived

508

from the many hadith samples that we studied carefully during this research. Experiments on sample hadiths show our approach to have a very good success rate. The work is far from over and there is a lot of room for improvement. We need to refine the grammar to be able to handle cases which it currently does not. For future improvement, we suggest handling the equivalent names problem. This is necessary to properly render the narration chain. We also need to modify the graphical rendering algorithm to use curved lines or angled lines to

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

link between different nodes. Of more practical improvement to the hadith scholars is to link the individual nodes in the graph with the narrator’s biographic database. A defective link in the isnad can be displayed using dotted lines as opposed to solid lines for unbroken links (see § Introduction).

REFERENCES Al-Osaimi, A., Al-Bassam, H., Al-Mazyad, K., Al-Yahya, K., Al-Ageel, M., Al-Rashed, R., & Al-Shalan, S. (2008). Al Asaneed tree. Retrieved November 15, 2009, from http://colleges.ksu.edu. sa/Arabic Colleges/ComputerScience/IT/DocLib/ BSC projects/Al Asaneed Tree.pdf. Al-Qushairy, M. H., & Siddiqi, A. H. (1972). Sahih Muslim: Being traditions of the sayings and doings of the Prophet Muhammad as narrated by his companions and compiled under the title AlJami‘-us-Sahih by Imam Muslim (Siddiqi, A. H., Trans.). Lahore, Pakistan: Sh. Muhammad Ashraf. Al-Salman, A. M., Al-Ohali, Y., & Al-Rabiah, M. (2006). An Arabic semantic parser and meaning analyzer. Egyptian Computer Science Journal, 28(3), 8–29.

Daelemans, W., & van den Bosch, A. (1992). Generalization performance of backpropagation learning on a syllabification task. In M. Drossaers & A. Nijholt (Eds.), Proc. 3rd Twente Workshop on Language Technology (pp. 27-37). Enschede, The Netherlands: Universiteit Twente. Daelemans, W., van den Bosch, A., & Weijters, T. (1997). IGTree: Using trees for compression and classification in lazy learning algorithms. In Aha, D. (Ed.), Artificial Intelligence Review: Special Issue on Lazy Learning (Vol. 11, pp. 1–5). Extended Backus-Naur Form. (n.d.). In Wikipedia. Retrieved November 12, 2009, from http://en.wikipedia.org/wiki/Extended_ Backus%E2%80%93Naur_Form. Habash, N., Soudi, A., & Buckwalter, T. (2007). On Arabic transliteration. In A. Soudi, A. van den Bosch & G. Neumann (Eds.), Text, Speech and Language Technology: Vol. 38. Arabic Computational Morphology: Knowledge-based and Empirical Methods (pp. 15-22). The Netherlands: Springer.

Arabic Transliteration. (2002). Qamus LLC website. Retrieved November 10, 2009, from http:// www.qamus.org/transliteration.htm.

International Standards Organization. (1996). ISO/ IEC 14977, Information technology – Syntactic metalanguage – Extended BNF. Retrieved November 12, 2009, from http://standards.iso.org/ ittf/PubliclyAvailableStandards/s026153_ISO_ IEC_14977_1996(E).zip.

Azami, M. M. (1977). Studies in Hadith methodology and literature. Indianapolis, IN: American Trust Publication.

Modern Standard Arabic. (n.d.). In Wikipedia. Retrieved April 9, 2010, from http://en.wikipedia. org/wiki/Modern_Standard_Arabic.

Azami, M. M. (1978). Studies in early Hadith literature (2nd ed.). Indianapolis, IN: American Trust Publication.

Naur, P. (1960). Revised report on the algorithmic language ALGOL 60. Communications of the ACM, 3(5), 299–314. doi:10.1145/367236.367262

Chomsky, N. (1956). Three models for the description of language. Information Theory. IEEE Transactions, 2(3), 113–124.

Peirsman, Y. (2006). Example-based metonymy recognition for proper nouns. Proc. 11th Conf. of the European Chapter of the Assoc. for Computational Linguistics (EACL-2006): Student Research Workshop (pp. 71-78). Trento, Italy.

509

Mining and Visualizing the Narration Tree of Hadiths (Prophetic Traditions)

Silberztein, M. D. (1997). The lexical analysis of natural languages. In Roche, E., & Schabes, Y. (Eds.), Finite-state language processing (pp. 175–204). MIT Press. Stanfill, C., & Waltz, D. (1986). Toward memorybased reasoning. Communications of the ACM, 29(12), 1213–1228. doi:10.1145/7902.7906 Stav, A. (2006). Shallow parsing. In Seminar in Natural Language Processing and Computational Linguistics. Tel Aviv Univ. Retrieved December 10, 2008, from http://www.cs.tau.ac.il/~nachumd/ NLP/shallow-parsing.pdf. Veenstra, J., & Buchholz, S. (1998). Fast NP chunking using memory-based learning techniques. In F. Verdenius & van den Broek (Eds.), Proc. of Benelearn’98 (pp. 71-78). Wageningen, The Netherlands.

510

Zavrel, J., & Daelemans, W. (1997). Memorybased learning: Using similarity for smoothing. Proc. 35th Annual Meeting of the Assoc. for Computational Linguistics (ACL-97) (pp. 43643). Madrid, Spain.

ENDNOTE 1



Anno Hegirae (in the year of Hijra) is the Muslim lunar calendar. In reference to the Prophet’s migration from Makka to Madina on 1 AH (corresponds to 621 CE).