ࡱ>   bjbj ӡJJ)))))t!*!*!**,!*Ж:.....z/z/z/KMMMMMM$ q)3z/z/33q))..,U,U,U3).).K,U3K,U,U2. 4BR!*Pې.70Ж S\)z/0,Ud12z/z/z/qq T z/z/z/Ж3333z/z/z/z/z/z/z/z/z/J S(: @ECorrelates between Performance, Prosodic and Phrase Structures in Bangla and Hindi: Insights from a Psycholinguistic Experiment Kalika Bali Monojit ChoudhuryMicrosoft Research Lab IndiaSadashivnagar, Bangalore 560080, India {kalikab,monojitc}@microsoft.com Diptesh Chatterjee1 Sankalan Prasad1Arpit Maheswari1Indian Institute of Technology KharagpurIndian Institute of Technology BombayKharagpur, West Bengal 721302, IndiaPowai, Mumbai 400076, India Abstract Shallow parsers or Chunkers that create word groups based on local dependencies are considered essential for many NLP related tasks such as Machine Translation, Speech Synthesis and Information Retrieval. The linguistic stature of chunks is accepted even though their definition varies with the linguistic theory, specific task and even practical concerns of system building. The psychological reality of a chunk in linguistic performance is under-explored for Indian languages. In this study we conduct a set of psycholinguistic experiments in Bangla and Hindi in an attempt to understand how linguistically structured chunks correspond to a nave native speakers perception of chunk in an attempt to understand the correlations between performance and linguistic structures. The results reaffirm the cognitive basis of chunks and show some interesting patterns in their correspondence with linguistic structure. This may have consequences for how speech synthesis models prosodic phrasing and also the computationally appealing notion of a local dependency based chunk. Introduction The grouping of words into chunks has been long accepted as a useful first step towards syntactic parsing in Natural Language Processing. Shallow parsers or chunkers that create word groups based on local dependencies are considered de rigueur for Machine Translation, Speech Synthesis and even Information Retrieval (Watanabe et al, 2003, Atterer, 2003, Liu et al, 2004). Production data of speech utterances in languages such as English supports the cognitive reality of such chunks (Selkirk 1983). Similarly, there is evidence from a number of psycholinguistic experiments for English that correlate perceptual data to corresponding linguistic structures that sentence processing by humans involves chunking (Gee and Grosjean, 1983, Abney, 1995). However, we know of no such attempt to explore the cognitive evidence behind the linguistically motivated chunks for Indian languages. Different linguistic theoretical frameworks and practical concerns can lead to different definitions, and hence, very different groupings of chunks. A chunk could thus be defined as a sub-tree within a syntactic phrase structure tree corresponding to Noun, Prepositional, Adjectival, Adverbial and Verb Phrases (Abney 1991, 1992, 1995) or as Noun Group and Verb Group based only on local surface information (Bharati et al 1995). However, as pointed out in Abney (1991), all linguistic models are nothing but an attempt to explain real use of language. Therefore, a theoretical linguistic unit of language cannot be independent of its perceptual reality. The validation of any linguistic theory by necessity lies in its ability to explain linguistic performance. In this study, we conduct a set of psycholinguistic experiments in Bangla and Hindi in an attempt to understand how linguistically structured chunks correspond to a nave native speakers perception of chunk as a word grouping. Corresponding prosodic data is also considered to find correlations between performance and linguistic structures. The analysis of the data from these experiments supports the cognitive status of chunks in Bangla and Hindi, and shows some interesting patterns in their correspondence with linguistic structure. The results have direct consequences for how prosodic phrases are modeled in speech synthesis for a more natural speech output. It also examines the different approaches towards chunking in NLP tasks and there basis in actual cognitive parsing by the native speakers. The rest of the paper is organized as follows: we discuss previous work done in related areas of psycholinguistics and computational linguistics in Section 2. Section 3 describes the experiment design and set-up in detail. The observations and results from the experimental data are discussed in Section 4. In Section 5 we present the implications of the observations that lead to certain conclusions. Definition of a Chunk One of the earliest attempts to define a linguistic chunk for a shallow parser is presented in Abney (1991). Abneys (1992) work follows from the sentence-parsing psychological experiments by Gee and Grosjean (1983). He replicates their experiment to study the performance structure of English as observed in nave parsing done by native speakers and its manifestation in corresponding speech data. Based on his experiments, Abney (1995) revised Gee and Grosjeans notion of a -phrase to define a chunk. For Gee and Grosjean, a -phrase with a boundary between a content word and the following function word, is best used to explain performance structure (nave parses). However, as Abney points out, they fail to adequately explain certain syntactic clusters. For example, in Fig. 1, the nominal modifier and noun would result in two separate -phrases, viz., [the old] and [man], in deviance from all experimental data. Abney proposes that chunks are connected sub-graphs of a parse tree not necessarily sharing constituency.  Fig 1: Chunks for English sentence (Abney 1992) where the chunks are marked with the brackets and the dependencies are shown through the arcs. Each chunk therefore, consists of a dominant head which in this particular example (Fig.1) is man. He goes on to show the syntactic validity of chunks as intermediate levels of analysis to a full parse structure. Abney (1995), thus, defines a chunk as the non-recursive core of an intra-clausal constituent, extending from the beginning of the constituent to its head, but not including post-head dependents. He lists seven chunk categories and introduces the notion of maximal chunk, i.e., a chunk that is not contained inside another chunk. It may be noted that this approach allows for the existence of certain words, for example, coordinators and subordinators, to not be a part of any chunk. Also, a chunk may contain elements that exhibit tight bonding based on morphology, syntax and semantics, a chunk is strictly a syntactic entity. Any work on defining chunks in order to create shallow parsers for Indian languages like Hindi and Bangla has to deal with the relative free word-order of these languages which allows for the movement of chunks within a sentence. NLP researchers have attempted to overcome such issues by defining chunks in Indian languages as having a fixed internal order with a sentence as bag of chunks that can move around in various permutations (Ray et al 2003). In the context of Indian languages, we find two very different approaches to chunking or grouping of syntactically related words. Bharati et al (1995) defines a local word group as a non-hierarchical sequence of words which can be formed based on local information, but are sufficient for further processing of the sentence according to the Paninian karaka theory. Examples include noun group (noun + postpositions) and verb group (verb + auxiliary). They justify this definition by claiming that (a) the concept of verb phrase does not seem natural for Indian languages, and (b) from a computational point of view, recognition of noun and verb phrases is hard. Nevertheless, for agglutinative languages such as Tamil, where the case-markers are affixed to the noun roots, and aspectual, tense and gender markers are affixed to verb roots, there seem to be no scope of local word grouping; almost all the words will remain ungrouped or equivalently, as singleton groups. Bharati et al (1983) suggest a rule-based algorithm for identifying local word groups for Hindi. In Sangal et al (2007), several research groups reported their experiments on using machine learning approaches for developing chunkers as a part of part-of-speech tagging and chunking competition, for which the chunking scheme was similar in spirit to that of Bharati et al. Das et al (2005) on the other hand introduced a concept of nested chunks. According to this approach, every word is a chunk to begin with; subsequently, two adjacent chunks can be joined to form a larger chunk. The process of chunking stops when a sentence is divided into multiple non-overlapping chunks all of which can be freely permuted within the sentence without changing the meaning. A chunk, therefore, is a sequence of words in a sentence which cannot be intruded by any word from outside it, but the words within a chunk and the chunks themselves are allowed to permute. They proposed a valence-theoretic rule-based algorithm for chunking. Ray et al (2003) suggested a chunking scheme which lies in between the two extremes. It is non-hierarchical in nature, but identifies simple phrases which are typically longer and more complex than local word groups. The algorithm proposed for chunking is again rule-based. Experiment Design In his experiments, Abney (1995) asked nave speakers of English to bisect sentences at their most natural break, and continue to bisect the resulting word clusters until they are left with a single word. Boundary strengths were assigned to each node based on the number of non-terminal nodes in that cluster, and the final boundary strength values were obtained by averaging across all speakers for each cluster. Corresponding speech data was also analyzed. Intra-sentential pause duration was used to mark major and minor prosodic boundaries to obtain prosodic phrases. The prosodic boundaries in this case were found to correlate very well with the boundaries of the nave parsing. In order to explore the cognitive reality of chunks for a shallow parse of a sentence in Bangla and Hindi we designed and conducted experiments similar to those of Abney (1995) for English. As has been stated earlier in the paper, we hope to achieve through these experiments a better understanding of how a native speakers linguistic performance relates to the theoretical linguistic structures. The Tasks The experiments were divided into two tasks a) text parsing and b) prosodic parsing. Each task was carried out by 6 native speakers in Bangla and Hindi. None of the speakers had any linguistic expertise but were fluent in reading, writing and speaking the language concerned. As in the earlier studies on English (Gee and Grosjean, 1983, Abney, 1995) the first task was aimed at eliciting performance structures from nave speakers. In this task, native speakers of the language were required to bisect sentences iteratively into their most natural groupings until the users felt that the word groups could not be divided any further. For example (see Table 1(b) for transliteration and gloss), [H5] (( , 0 )(8 A ( $ G 9 @ ))((. H  )($ A  0 $ )(( 0 8 G )(- >  > ))) All the sentences were also recorded in a separate session as read out by each speaker to elicit natural prosodic breaks. The task was carried out in both Hindi and Bangla. The speech recordings were made on a high quality PCM digital recorder in a normal office environment and the data analysed using the speech analysis software PRAAT (Boersma et al 2009). Choosing the Sentences The sentences for testing in both Bangla and Hindi are based on certain common structures of Indian languages. We sought to test through these sentences not only the role word-order plays in the perception of a chunk by nave users, but also certain syntactic constructions where chunking is not always straightforward. A set of 10 sentences for Bangla and 9 for Hindi were used to specifically cover the following syntactic structures: Embedded clauses Sentence and phrase-level adverbs Conjuncts Relative clauses Participles Noun Groups: Compound Nouns, Named Entities and Multiword Expressions (multiple) Adjectives+ Noun Qualifier + Adjectives + Determiner + Noun Noun + Postpositions Complex Postpositions Verb Groups: Compound Verbs (Polar+Vector) Verb + (multiple) Auxiliaries + Particles (aspect/negation/wh-) Noun + Verb (group) Table 1, at the end of the paper, includes all the sentences used for a) Bangla and b) Hindi. The sentences in two languages are near translations to make a comparative analysis easier. Observations The results from both the text parsing and the prosodic parsing experiments are presented in Table 1 (a) and (b). For every example, first line shows the example sentence, followed by its transliteration into roman, gloss and English translation. The transliteration has been annotated with POS tags (marked as word/TAG) and multiword expressions (marked by []TAG). See Table 2 for description of the POS and multiword expressions tags. The next two lines show the boundaries obtained from prosodic parsing and text parsing, respectively. The representation convention is as follows: A hyphen denotes a word. The order of words in the original sentence is maintained. Note that hyphenated words have been considered as a single word and punctuation marks have been ignored. The breaks are marked as major |, and minor *| (see Section 4.1, 4.2 for explanation of major and minor breaks) Prosodic-Parsing Experiment Waveforms of each recorded sentence were analysed and intra-sentential pause durations labeled. The durations thus extracted were normalized for each speaker to rule out effects of reading rate and speaker variation. The pause durations fell into two distinct classes: short pauses of <4.5 ms and long pauses of >7 ms. A minor boundary, marked with an *, was posited at each short pause and a major boundary at each long pause. The number of speakers who produced that pause was also enumerated at each boundary. Thus, [B5] - 2*| - 6| - - - 6| - - - 6| - - - implies that 2 of the subjects have a minor break after shimlA while all the subjects have a major break in the other places marked. Again, as in the text parsing experiment, there is very good correlation within subjects. Some points to note for both Bangla and Hindi: There is always a major break at clause boundary, and major prosodic boundaries also correspond very well with the phrase structure of the sentence. The sentential adverb is consistently separated by major boundaries by all speakers in both languages (B1, H1) In general, postpositions are chunked with nouns. A consistent exception is H4 where sAhitya akAdemI and ne have a major boundary in between. Similarly, compound nouns are always chunked together except trinamUl kAn.gres in H1 the two nouns by half the speakers. This might be because of unfamiliarity of the speakers with the compound itself. The classification of coordinating and subordinating particles as not being part of any chunk (Section 2) is supported by this data as these particles are sometimes grouped with the preceding and sometimes with the following clause. There also is some role of focus in determining the prosodic breaks. For example, in B7 laal, neel, sabuj pataaka duliye duliye was chunked differently based on whether the subject wanted to emphasise the colours of the flags or the act of waving them. Text-Parsing Experiment For the text-parsing experiment, we asked nave native speakers of the language to construct a binary tree out of the sentence by appropriate bracketing, such as: [B3] ( )(( ( ))(( )( ))) Then the brackets were numbered according to their height in the underlying binary tree. We show it for the above example where the bracket numbers are presented as subscripts. (0 )0 (0 (1 (2 )2 )1 (1 (2 )2 (2 )2 )1 )0 Between every pair of words, the break is assigned a weight equal to the minimum of the closing or opening brackets at that point. The weight of a break between a word pair which are adjacent (i.e., the subject has not introduced any brackets between the words) is assigned a score of 1 more than the minimum of the nearest brackets. Thus, we get 0 1 3 1 3 2 3 The boundary scores are then averaged over 6 speakers. Let us assume that the above example shows the average score. The boundary score at each word is subtracted from the maximum average boundary score over all boundaries (which is equal to 3 for our running example). This is done in order to assign a higher score to a stronger boundary. This gives us 3 2 0 2 0 1 0 The scores thus obtained are then converted to a fraction of the maximum boundary score (i.e., 3, in this example). Hence, 1 0.67 0 0.67 0 0.33 0 We consider a score over 0.67 as a major boundary and one between 0.33 and 0.67 as a minor boundary. Therefore, | *| *| Final structure: - | - *| - - *| - - - - Both the languages showed a good agreement amongst the subjects and it was observed that: Noun is grouped with the following postpositions. H4 shows an exception where a relative clause is embedded between the Named Entity premchan.d and the postposition ke. This holds true for spatio-temporal nouns as well (B10). Named entities and MWE are always grouped together (B1, H1). Adjectives, including multiple adjectives, are grouped with the noun (B7). However, for qualifier + adjective + determiner + noun, even though all subjects agree that these should be chunked together (H9), there is a lot of ambiguity on how to group words within the chunk. Fig. 2 illustrates this ambiguity through the groupings obtained for H9 from the 6 subjects. Complex postpositions have been chunked into a single group and then chunked with the previous noun. (B10, H7) There is high consistency in the verb chunks as polarsvectors, aspect markers, negatives and even question markers are consistently chunked with the main verb. However, there is a difference in case of negation. (for example, H7 versus H2, H3, and B2 versus B5, B6) For participles (B4) while there is a clause-level agreement, there is some confusion in chunking at the lower levels. Sentential adverbs (B1, H1) while chunked as a part of the following Noun chunk, are marked separate by a minor break in both the languages.  Fig 2: Various groupings of a noun phrase from H9 by the native speakers show that there is ample ambiguity in the perception of the chunks within a noun phrase. Comparing Prosodic Structure to Performance Structure The data from the two parsing experiments showed good correspondence between the prosodic structure and the performance structure, at least for the major boundaries. As the number of minor boundaries in the prosodic data was comparatively few and also as it was not possible to analyze the prosodic structure as hierarchical in nature, it is difficult to say much about the correspondence between the minor boundaries in the two datasets. However, if we look at comparable sentences, B1 and H1, for example, we find that the breaks are quite consistent. A few anomalies can be observed between the prosody and text data. For example, B4, where the large number of prosodic breaks are not reflected in the text grouping. Though this cannot be conclusively determined from the current data, it is our intuition that the prosodic chunks are not entirely determined on the basis of syntactic structure alone but are influenced by phonological considerations like the number of syllables in a chunk and discourse-level phenomena such as topic and focus. Discussion and Conclusion In this study, we have made an attempt to understand the nature of chunks in Bangla and Hindi from a cognitive perspective. We have observed that the prosodic and performance structures agree with each other, which in turn matches with the phrase structure of the sentence. Moreover, there is a good amount of agreement between the speakers on the major prosodic and performance breaks. Even though the experiments have been conducted with a small set of sentences and few speakers, the systematic and consistent behavior of the three kinds of structures across the speakers and the languages makes us believe that it is possible to make some significant generalizations from the observations. Firstly, we would like to highlight that chunking, at least from a cognitive perspective, is a sentence level phenomenon and therefore, local information is not sufficient for obtaining chunks. For instance, a very commonly used local rule for chunking is to introduce a chunk boundary after a postposition. However, we note significant deviations from this rule such as: [H1]: No chunk boundary after kAn.gres ke, because kAn.gres ke sadasyoM ne forms noun a phrase. [H3]: No prosodic boundary after shimlA se, because shimlA se manAli is being treated as a phrase. This is more evident when we compare H5 and H8. While in the former case the verb bhAgA forms a chunk (verb phrase) with the preceding postpositional phrase ghar se, in the latter case it forms a chunk with the adverb turant. This suggests that there is an underlying notion of verb phrase in Indian languages, even though its syntactic structure is largely governed by context. Bangla being a mildly agglutinative language, we do not find such instances. Nevertheless, while in B2 we observe a minor break between the genitive Taber and the modified phrase bhA.nga TukRo, no such break is observed in B1 between the genitive ka.ngreser and the modified sadasyarA. This is due to the presence of the adjective bhA.ngA in the former case, and therefore, indicates that chunk boundaries might also depend on factors such as length of the chunk and the distance between the modifier and the modified. Secondly, we want to reemphasize the fact that while speakers seem to strongly agree on major breaks that correlate well with linguistic definition of phrases and clauses, there is not much agreement when it comes to chunks within phrases. For instance, in B2 the speakers have grouped EkTa bhAlo wAl haen.gin.g variously, and therefore upon averaging we observe no breaks, though in the prosodic structure there is a major break after bhAlo. Similar performance structure can be observed for Hindi (H1). Thus, it is important that any chunker for speech applications makes use of global information to identify clause boundaries as well as major phrase boundaries. We also note that while the concept of local word grouping (Bharati et al 1995) is computationally appealing, there is no consistent correlation between such word groups and prosodic or performance structures. On the other hand, clause and phrase level groupings as suggested by (Das et al, 2005) seem to have a cognitive underpinning. However, it should be noted that while the scheme proposed by Das et al. is hierarchical, our analysis, by definition cannot elicit hierarchical structures out of the data. References A. Bharati, V. Chaitanya, and R. Sangal. 1995. Natural Language Processing: A Paninian Perspective. Prentice Akshay Singh, Sushma Bendre, and Rajeev Sangal. 2005. Hmm based chunker for Hindi. In Proceedings of IJCNLP-05. Jeju Island, Republic of Korea. Dipanjan Das, Monojit Choudhury, Sudeshna Sarkar and Anupam Basu. 2005. An Affinity based Greedy approach towards Chunking for Indian Languages. Proceedings of the International Conference on Natural Language Processing (ICON), pp. 55 - 62, Kanpur, India Elisabeth O. Selkirk. 1984. Phonology and Syntax: The Relation between Sound and Structure. The MIT Press, Cambridge, MA. Feifan Liu, Qianli Jin, Jun zhao, and Bo Xu. 2004. Bilingual Chunk Alignment Based on Interactional Matching and Probabilistic Latent Semantic Indexing. In Proc. of IJCNLP, 2004. Gee, James Paul, and Francois Grosjean 1983. Performance Structures: A Psycholinguistic and Linguistic Appraisal, Cognitive Psychology 15, 411458. Michaela Atterer. 2002. Assigning prosodic structurefor speech synthesis: a rule-based approach. In Proc. of the Speech Prosody 2002 Conference, Aix-en-Provence. Paul Boersma, and David Weenink, 2009. Praat: doing phonetics by computer (Version 5.1.12) [Computer program]. Retrieved August 4, 2009, from  HYPERLINK "http://www.praat.org/" http://www.praat.org/ P. R. Ray, Harish, V., S. Sarkar, and A. Basu. 2003. Part of Speech Tagging and LocalWord Grouping Techniques for Natural Language Parsing in Hindi. Proceedings of International Conference on Natural Language Processing (ICON 2003). Mysore, India. Rajeev Sangal, Sushma Bendre, Dipti Misra Sharma, Prashanth Reddy (ed.) 2007. Proceedings of the IJCAI 2007. Workshop on Shallow Parsing for South Asian Languages ( SPSAL-2007 ) Hyderabad, India Steven Abney. 1991. Parsing by Chunks. In: Robert Berwick, Steven Abney and Carol Tenny (eds.), Principle-Based Parsing. Kluwer Academic Publishers, Dordrecht. Steven Abney. 1992. Prosodic Structure, Performance Structure and Phrase Structure. In: Proceedings, Speech and Natural Language Workshop, pp.425-428. Morgan Kaufmann Publishers, San Mateo, CA. Steven Abney.1995. Chunks and Dependencies: Bringing Processing Evidence to Bear on Syntax. In: Computational Linguistics and the Foundations of Linguistic Theory. CSLI. Taro Watanabe , Eiichiro Sumita , Hiroshi G. Okuno. 2003. Chunk-based statistical translation, Proceedings of the 41st Annual Meeting on Association for Computational Linguistics, p.303-310, July 07-12, 2003, Sapporo, Japan Table 1 (a) Text and Speech Parsing data in BanglaEx#Bangla sentences and the chunk boundariesB1 gatakAl\RB [mahatma\NP gAndhi\NP rode\NP]NE [triNamul\NP kan.greser\NP] NE sadasyarA\NN Ek\DT viSAl\JJ patha\NN avarodher\NN [Ayojan\NN korechhilo\VM]CV yesterday Mahatma Gandhi Road-loc Trinamool Congress-gen members a huge road block-gen organize do-past Yesterday, the members of Trinamool Congress organized a huge roadblock on Mahatma Gandhi Road. - 6| - - 2| - 6| - - 2| - 6| - 2*| - 2| - - 3| - - - *| - - *| - | - - *| - | - - - - *| - - B2 , Tober\NN bhAn.gA\JJ Tukrogulo\NN phele\VM diyo\VA nA\RP, ogulo\PR diye\PP ekTa\DT bhAlo\JJ [wAl\NN hyan.gin.g\NN]CN bAnAno\VN jete\VA pAre\VA Pot-gen broken piece plural throw give no those using one good wall hanging create go can Dont throw the broken pieces of the pot; a good wall-hanging can be made out of those. - 2*| - - - 5| - - - 6| - - 4| - - 5| - - 6| - - - - *| - - - | - - *| - | - - *| - - - - | - *| - - B3 ? Tumi\PR rAmke\NP lAThi\NN diye\PP eto\JJ jore\RB mArle\VM kEna\PR you Ram-acc stick with so hard beat-past why Why have you beaten Ram with a stick so hardly? - 1*| - 6| - - 6| - - 6| - - - | - *| - - *| - - - - B4 , , [jibanAnanda\NP dAs\NP]NE, jinni\PR [baNalatA\NP sen\NP] NE likhe\PL [vikhyAta\JJ hayechhilen\VM] CV, onAr\PR prati\PP [shraddhA\NN jAnate\PL] CV [bAn.glA\NP AkAdemi\NP] NE ekTA\DT natun\JJ puraskAr\NN [ghoSanA\NN korechhe\VM] CV Jibananada Das who Banalata Sen write-participle famous be-past his towards respect show-inf Bangla Academy a new award announcement do The Bangla Academy has announced a new award in honour of Jibananda Das who is well known for composing Banalata Sen. - - 6| - *1| - 4| - 1| - 3| - 2| - 2| - 1*| - 1*| - 1| - 2| - 2*| - 6| - 2| - 2| - 3| - 4*| - - - | - - - - - - | - - - - - - - - - - - B5 shimlA\NP haye\PP mAnAli\NP nA\RP giye\PL sojA\RB dilli\NP thekei\PP flAiT\NN niye\VM jAn\VA Shimla via Manali no go-participle straight Delhi from flight take go Rather than going to Shimla via Manali, take a flight directly from Delhi. - 2*| - 6| - - - 6| - - - 6| - - - - - *| - *| - - | - *| - - *| - - - B6 , Ami\PR bhebe\PL rekhechhi\VA pujAr\NN Agei\NST ei\DT kAjTa\NN [sheS\JJ kare\VM]CV phelbo\VA, kAroN\CNJ pare\RB Ar\JJ samay\NN pAba\VM nA\RP I think-participle have prayer-gen before this work finish do throw-fut because later more time get-fut no I have decided to finish this work before the prayer because I will not find time later. - - - 6| - - 6| - - 5| - - - 6| - 6| - 1*| - - - - - *| - - | - - *| - - *| - - - | - | - *| - *| - *| - - B7 , , lAl\JJ, nil\JJ, sabuj\JJ patAkA\NN duliye\PL duliye\PL bAchchArA\NN bhAratiya\JJ khElowARder\NN [swAgata\NN jAnAlo\VM] CV red blue green flag waive-participle waive-participle children Indian players-gen welcome show-past Children welcomed the Indian players by waiving red, blue and green flags. - 6| - 6| - 3| - 3| - - 6| - 4| - - 2| - - - - - - *| - - | - *| - - *| - - B8 , driR\JJ, athacha\CNJ mridu\JJ kanThe\NN tini\PR Ek\DT atyanta\JJ preranAdAyak\JJ [vaktritA\NN diyechhilen\VM] CV firm but soft voice-loc he a very motivating speech give-past In a firm yet soft voice he delivered a very motivating speech. - 3| - 3| - - 6| - 2| - 3| - 2| - 4| - 3| - - *| - - - | - *| - - - - - B9 ? Tumi\PR rAmke\NN Eto\JJ jore\RB lAThi\NN diye\PP mArle\VM kEno\PR you Ram-acc so had stick with beat-past why Why have you beaten Raz{   $ % * 蹣xixixTix<.hCTh3:95CJH*OJQJ^JaJnHtH(hCTh3:9CJOJQJ^JaJnHtHhCTh3:9CJaJnHtH(hCTh3:9CJOJQJ^JaJnHtH+hCTh3:95CJOJQJ^JaJnHtH+hCTh3:95CJOJQJ^JaJnHtH(hw5CJOJQJ\^JaJnHtH2h@h@5CJNHOJQJ\^JaJnHtH.h@h@5CJOJQJ\^JaJnHtHWAkdJ$$IfF## t644 laytCTAkd$$IfF## t644 laytCT$$7$8$H$Ifa$gdCTl  $7$8$H$a$gdw   > O cAkd$$IfF## t644 laytCT$$7$8$H$Ifa$gdCTl Akd$$IfF## t644 laytCT* - < = > M N O P x y F G L M   ӻӻӗӗ{t{i\i\i\i\i\i\i\i\ih3:9h3:9NHnHtHh3:9h3:9nHtH h[hw h)Fhw(hCTh3:9CJOJQJ^JaJnHtHhCTh3:9CJaJnHtH(hCTh3:9CJOJQJ^JaJnHtH.hCTh3:95CJH*OJQJ^JaJnHtH+hCTh3:95CJOJQJ^JaJnHtH+hCTh'5CJOJQJ^JaJnHtH$O P y =Tkd$$IfF0F#I t644 laytCT$$7$8$H$Ifa$gdCTl Tkd($$IfF0F#I t644 laytCT "/T\rzzzzGgdIgdgd 2 & F^gdw & Fgdwgdw`gdwTkd$$IfF0#@ t644 laytCT !"/JKP"HIv{uvdeƹƝƝƍƝ}p}cƝƝƝƝƝƝchxjhPJnHtHh fMhPJnHtHh fMh6PJ]nHtHh$h6PJ]nHtHhNHPJnHtHhxjh6PJ]nHtHh6PJ]nHtHhPJnHtH h1hwh3:9hwnHtH*jh3:9h3:90JB*UnHphtHh3:9nHtH&)*@AFKMN+,:;vwXY-.TUqrdehjD꺲hnHo(tHhNHnHtHhnHtHhhPJnHtH$hdh56PJ\]nHtHh_hh6PJ]nHtHhPJnHtHhNHPJnHtH>DFtv ) * + , ; [ !!!!!!!!!!zogogo\h]c^hnHtHhnHtHh]c^hnHtHhgnHtHh]hnHtHh_hhCJnHtHhi-hCJNHnHtHhCJnHtHhi-hCJnHtHht<{hnHtHjHht<{hUnHtHh 9h 9nHtHhnHo(tHhnHtHhNHnHtH ) * , ! $%+-..r13 3"455\6^6889d:gdgdGgdRgdIgdI$a$gdgd 9I`gd!!!"""""#"$"Q"R"Y"Z"""""!#"#R#S#$$ $ $$$$$%%%%%]&b&d&h&k&l&('௥|||m|h{khNHPJnHtHh{khPJnHtHhC9dh6PJ]nHtHhNHPJnHtHhPJnHtHhLy hnHtHh=h6]nHtHhNHnHtHhOJQJ^JnHtHhnHtHh]c^hnHtHh]c^hNHnHtH)('+','>'C'D'b''''''(($(%(A(B(|(}((((( ))&),)N)O)~))))))))))) *****N*O****** ++++++"+p+ѸѸѸѸѸѨѸѸѸѸѸѸѸhI`h6PJ]nHtHhC9dh6PJ]nHtHhPJnHtHh{khNHPJnHtHh{khPJnHtH"h0h6NHPJ]nHtHh0h6PJ]nHtH;;;;;;;;;;;Q<R<{<<<<<<<<<<$=%=4=M====ʽyoehgPJnHtHhPJnHtHh$-hPJnHtHhxYhPJnHtHhNHPJnHtHhI!hNH hI!h hd'h h8Hhh{gqhNHmH sH h{gqhmH sH jh0JUh<hPJnHtH hNHhPJnHtHh'd:u::::::; ;K;`;v;;;;;;<<>?(??B@LgdRgdIgdgdGgdgdI & FgdI & Fgd====>>.>1>3>5>F>G>\>v>|>}>>>>>>>>>>>>M?N?!@"@+@,@A@B@@@BBBBB{h$MhPJ_HEnHtHhr=hPJnHtHhNHPJnHtHh$-hNHPJnHtHh$-hPJnHtHhgNHPJnHtHhH*PJnHtHhNHPJnHtHhgPJnHtHhPJnHtHhPJnHtH)B@^@8ABBDCpCCCnDDDEEF[GDHDI\IIJrJKNgdG & FgdgG & FgdGgdR`gdgR`gdRgdIgdgdBBBBBBBBCCCCCC&C(C2C4C@CBCXCbCdCjClCnCpCCCCCCD%DFDGDmDnD#EҾҾҾҾҾҾҾҾҾҾҾҫwmchPJnHtHhNHnHtHhi-h6]nHtHh!8hnHtHhgnHtHh*hnHtHhnHtH$h$hCJPJ_HEaJnHtH'h$hCJPJ_HEaJnHtH6h$hCJOJPJQJ^J_HEaJnHo(tH"hOJPJQJ^J_HEnHtH&#E$EPEQEEEEESFbFiFkFFFFFFFFFTGUGGGHHoHpHHHHH8I9ICIDI\IIIIIIIJ J Jʽʽꦗy h$hCJPJaJnHtHh$hPJnHtHhi-hNHPJnHtHhi-hPJnHtHhgPJnHtHh6PJ]nHtHhi-h6PJ]nHtHhGh6PJ]nHtHhPJnHtHhNHPJnHtH- JJJ$J(J0J2JW?W@WWW#X$XQXWXXXXXXXY YY裗wj[jh3"hNHPJ nHtHh3"hPJ nHtHhK<NHPJnHtHh$PJnHtHhK<PJnHtHhNHPJnHtHhPJnHtHhgPJ_HEnHtH"h3"h6PJ]_HEnHtHhNHPJ_HEnHtHhR&AhPJ_HEnHtHhPJ_HEnHtHh$PJ_HEnHtH"@WXYY+ZZ/[0[2[3[[[ \7^'`A`bodd:eN & FgdNgdgdGgdG`gdgdRgd$ R$`a$gd$R`gd$ Rh^h`gd$R & FgdYYOYYYYYZZZZZZZZ.[/[0[1[3[8[:[[[[[[[[[ \j\òxnbnZhnHtHhNHPJnHtHhPJnHtHh3"h$PJ nHtHh$CJNHnHtHhi-h$CJnHtHh$CJnHtHh$PJnHtH!j?hgh$PJUnHtHh$PJ nHtHh3"hNHPJ nHtHh3"hPJ nHtHh3"hPJnHtHhPJ nHtHj\k\\\0]1]f]g]]]%^*^-^.^^^________!`"`'`o`p`WaXaaabbbb'c(cccOdPdddeeeeffgfnfofffKgLg{g|ggg?h@hVhWhiiijajbjjjܶh|#hPJnHtHhr0hPJnHtHhNHPJnHtHhPJnHtHh$nHtHhnHtHhNHnHtHH:efhjjmummrnroop4qqrs_tuumvMwNwOwPwQwS$a$gdgl 3bJgdSgdPgdGgdG`gdjUkVkkkkkkklllllllmjmsmtmummmmm8nCnDnPnqnrnnnnno4o5oTorooooʽvlh{gqh6]h6NH] h{gqh hNH h6]h IhPJnHo(tHhPJnHo(tHhh fMhPJnHtHh6PJ]nHtHh fMh6PJ]nHtHh_hh6PJ]nHtHhNHPJnHtHhPJnHtH*oo p!pppppppppppqqq3qgqhqqqqqdrerrrrrrss6s8s9sDsEs|s}sssһҭҢvlh{gqh6]h6NH] h6]hRRh0JjWhUjhUh hnHtHh fMh6]nHtHhNHnHtHh{gqhNHnHtHhnHtHh{gqhnHtHh fMh6] hNHh h{gqh)ssssssssstttt4t5t>t?tMtNtStTt^t_tqtrtstttttttttuuuuuuuuvvo$h_hNH_H9mH @nH @sH @tH @ h_h_H9mH @nH @sH @tH @ h h_H9mH @nH @sH @tH @h_H9mH @nH @sH @tH @ hr|h hNH h6] hPJ ] h[h hoJ`h h{gqhh hNHh{gqh6]h h h+v#v%v&vfvlvmvvvvvvw w>wLwMwNwwww﷦|ndWGh)>hgCJaJnHo(tHhZhgPJnHtHhgPJnHtHhg_H9mH @nH @sH @tH @*h fMh6NH]_H9mH @nH @sH @tH @&h fMh6]_H9mH @nH @sH @tH @ h fMh_H9mH @nH @sH @tH @h_H9mH @nH @sH @tH @*h h6NH]_H9mH @nH @sH @tH @&h h6]_H9mH @nH @sH @tH @ h h_H9mH @nH @sH @tH @QwRwwwwwkXXS$Ifgdgl ;0b~kd<$$IfTl$$  t 0$644 lap ytgTS$$Ifa$gdgl 3bwwwwwwwx x xxx(x*x2x4x@xBxTxVxdxfxjxlxvxxx|x~xxxxxxxxxxxxxxxyyy y yyyݷ|h 9CJH*_HEaJh 9h 9CJH*_HEaJh 9CJ_HEaJh)>hgCJ_HEaJh)>hgCJaJ*h)>hgCJOJQJ^J_HEaJo(h)>hgCJaJnHo(tH#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtH0wwwxy[zzcP::::S$Ifgdgl m$S$Ifgdgl kd$$IfTl0$&"  t0$644 lapytgTy$y6y:yByFyHy\ybyhynytyvyzyyyyyyyyyyyyyyyzz{{{{{|ߩvh)>hgCJaJnHo(tH#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtHhgCJPJaJnHtHh)>hgCJNH_HEaJh 9CJH*_HEaJh 9h 9CJH*_HEaJhgCJ_HEaJh)>hgCJ_HEaJh 9CJ_HEaJ#zzz{{{{|}M~eRS$Ifgdgl kdΌ$$IfTl0$&" t0$644 laytgTS$Ifgdgl m$ || |||"|$|,|.|6|8|@|B|F|J|T|V|`|b|j|l|t|v|||||||||||||||||||} }}}} }"}(}6}<}F}L}V}X}\}h}hCJPJ_HEaJnHtHh 9CJPJ_HEaJnHtHhgCJPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH h)>hgCJPJaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH8h}n}p}r}x}~}}}}}}}}}}}~~  ͪ}kO6h)>hgCJOJPJQJ^J_HEaJnHo(tH#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tHhgCJaJnHtHh)>hgCJaJnHtH'h 9h 9CJH*PJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtHhgCJPJ_HEaJnHtHh 9CJPJ_HEaJnHtHM~~~~  S@S$Ifgdgl kd$$IfTl0$&" t0$644 laytgT$Ifgdgl m$S$Ifgdgl m$  *,02:<FHNPRTV\bntƀЀրځ ϼ~~~~~~~~oh)>hgCJaJnHtHhqCJPJaJnHtH h)>hqCJPJaJnHtHhgCJPJaJnHtH h)>hgCJPJaJnHtH$h)>hgCJPJ_HEaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH'h)>hgCJPJ_HEaJnHtH(T؀:ځ UkdB$$IfTl0$&" t0$644 laytgT$Ifgdgl S$Ifgdgl  &(.2:<FHNPXZhjz~̂΂ւ؂ (.̰̰yyhqCJPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH'h)>hgCJPJ_HEaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tH#h)>hgCJPJaJnHo(tH/nlm̅$Ifgdgl m$S$If^`gdgl m$S$Ifgdgl  .6>BFVXZjpxʃ҃ԃ؃ "%-14569<BENQRSZhgCJPJ_HEaJnHtH!hCJH*PJ_HEaJnHtH'hqhCJH*PJ_HEaJnHtHhCJPJ_HEaJnHtH'hqhqCJH*PJ_HEaJnHtHhqCJPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH4Z]fjkmlm̅ ɸݙzhXh<6h)>hgCJOJPJQJ^J_HEaJnHo(tHh)>hgCJaJnHo(tH#h)>hgCJPJaJnHo(tHhgCJPJaJnHtH h)>hgCJPJaJnHtHh)>hgCJaJnHtHhgCJPJ_HEaJnHtH!hCJH*PJ_HEaJnHtH'hqhCJH*PJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtHhCJPJ_HEaJnHtH:Ƈ\^}jLLLLLS$If^`gdgl m$S$Ifgdgl kd$$IfTl0$&" t0$644 laytgT "$(*46>@LNXZfhrtz|~†̆҆Ԇ܆$*28\^ϼhgCJaJnHtHh)>hgCJaJnHtHhgCJPJ_HEaJnHtHhCJPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH'h)>hgCJPJ_HEaJnHtH2^iV8S$If^`gdgl m$S$Ifgdgl kd$$IfTl0$&" t0$644 laytgT$Ifgdgl m$ "*,02<>DFLNX\dflnrt|~‰ȉԉډ ̰ްްްްްްްްްްްްްްްްhCJPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tH7 #&'*.369<>?BEJMPSɴɴɕyhV#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tHhgCJaJnHtHh)>hgCJaJnHtHhgCJPJ_HEaJnHtH(h)>hgCJNHPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH'hhCJH*PJ_HEaJnHtHhCJPJ_HEaJnHtHTLKkd$$IfTl0$&" t0$644 laytgT$Ifgdgl m$S$If^`gdgl m$\L$Ifgdgl m$S$If^`gdgl m$S$Ifgdgl   (*68DFVXfhʌЌތ &:@X^`bjlpv'hhZCJH*PJ_HEaJnHtH(h)>hgCJNHPJ_HEaJnHtHhZCJPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH h)>hgCJPJaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH0ď̏Ώҏԏ "$&06HP\bpv²qqqqqqqqqaaaaahwCpCJPJ_HEaJnHtH'h)>hgCJPJ_HEaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tHhgCJaJnHtHh)>hgCJaJnHtH$h)>hgCJPJ_HEaJnHtHhgCJPJ_HEaJnHtH&( }jLLLLLS$If^`gdgl m$S$Ifgdgl kd"$$IfTl0$&" t0$644 laytgTȐʐ̐АҐؐܐ͸͕yhL6h)>hgCJOJPJQJ^J_HEaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tHhgCJaJnHtHh)>hgCJaJnHtH'hhwCpCJH*PJ_HEaJnHtH(h)>hgCJNHPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtHhgCJPJ_HEaJnHtHhwCpCJPJ_HEaJnHtH`iV8S$If^`gdgl m$S$Ifgdgl kd$$IfTl0$&" t0$644 laytgT$Ifgdgl m$ȒʒҒԒޒ &06BHPRXdjtz02뾰uh)>hgCJaJnHo(tHhgCJaJnHtHh)>hgCJaJnHtHUhgCJPJaJnHtHhwCpCJPJaJnHtH h)>hgCJPJaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH'h)>hgCJPJ_HEaJnHtH.|ԓ02lKkd$$IfTl0$&" t0$644 laytgT$Ifgdgl m$S$If^`gdgl m$m so hard with a stick? - 1*| - 6| - - 6| - - 6| - - - | - | - - *| - - *| - - B10 , Megher\NN anek\JJ upar\NST diye\PP uRte\PL uRte\PL jakhan\RB klAnta\JJ haye\VM paRechhilo\VA bAjpAkhiTa\NN, takhanai\RB haThAt\RB nicher\NST theke\PP EkTa\DT AwAj\NN sunte\VM pelo\VA cloud-gen lot up through fly-participle fly-participle when tired be-participle fall-past hawk then suddenly down-gen from a sound hear get-past When the hawk got tired flying way above the clouds, suddenly it could hear a sound from below. - 4| - 1*| - - 6| - - 6| - 2*| - - - 4| - 6| - 1*| - 4| - - 4| - - 3| - - - *| - - - | - - | - *| - - - *| - | - *| - | - - *| - - *| - - Table 1 (b) Text and Speech Parsing data in HindiEx#Hindi sentences and the chunk boundariesH1 2 . 9 > $ M . >  >  ' @ 0 K ! * 0 $ C # . B 2  >   M 0 G 8  G 8 & 8 M / K  ( G   5 ? 6 > 2 0 > 8 M $ > 0 K  K 0 H 2 @  >  / K  (  ? / > | Kal\RB [mahatma\NP gA.ndhI\NP roD\NP] NE par\PP [triNamUl\NP kAn.gres\NP] NE ke\PP sadasyo.n\NN ne\PP ek\DT vishAl\JJ [rAstA\NN rook\NN] CN rElI\NN kA\PP [Ayojan\NN kiyA\VM] CV yesterday Mahatma Gandhi Road on Trinamool Congress of members ergative a huge road block rally of organize do-past Yesterday, the members of Trinamool Congress organized a huge roadblock rally on Mahatma Gandhi Road. - 6| - - 4*| - - 6| - 3| - - - 1*| - 5|* - - 5*| - - 5*| - - 6| - - - *| - - - - | - - - - - | - - - - - - - - H2 . 2 G  G  A  ! K   K + H  . $ & G ( > | gamle\NN ke\PP TukRo.n\NN ko\PP phE.nk\VM mat\RP denA\VA pot of pieces acc throw not give Do not throw away the pieces of the pot. - - 4| - - 6| - - - - - *| - - | - | - - H36 ? . 2 > 8 G . ( > 2 @ ( >  >  0 8 @ ' G & ? 2 M 2 @ 8 G 5 ? . > ( 2 G 2 K | shimlA\NP se\PP manAlI\NP nA\RP jAkar\PL sIdhe\RB dillI\NP se\PP vimAn\NN le\VM lo\VA Shimla from Manali no go-participle straight Delhi from plane take take Rather than going to Shimla via Manali, take a flight directly from Delhi. - - - 3| - - 6| - - - 6| - - - - - *| - *| - - | - *| - - | - *| - - H4* M 0 G .   & ,  ? (  @ * M 0 8 ? & M ' 0  ( >   . G   K & > ( * M 0 . A  9 H ,  G 8 . M . > ( . G  8 > 9 ? $ M /   > & . @ ( G   (  * A 0 8 M  > 0  @  K 7 # >  @ 9 H | premchand\NP, jinkI\PR prasiddh\JJ rachnAo.n\NN me.n\PP godAn\NP pramukh\JJ hE\VM, ke\PP sammAn\NN me.n\PP [sAhitya\NP aKAdamI\NP] NE ne\PP ek\DT nae\JJ puraskAr\NN kI\PP [ghoSaNA\NN kI\VM] CV hE\VA Premchand whose famous creations in Godaan main is of honour in Sahitya Academy ergative one new award of announcement do is Sahitya Academy has announced a new award in honour of Premchand, whose famous creations include Godaan. - 6| - - - - 6| - - - 6| - - - 6| - - 3| - 6| - 2*| - - - 6| - - - - | - *| - - - *| - *| - - | - - - | - - - | - - - *| - - *| - -H5 , 0 8 A ( $ G 9 @ . H  $ A  0 $  0 8 G - >  > | khabar\NN sunte\PL hI\PR mE.n\PR turan.t\RB ghar\NN se\PP bhAgA\VM news hear-participle only I immediately home from run-past I ran from home as soon as I got the news. - - - 6| - - 6| - - - - *| - - | - - *| - - - H6, > & 2 K   G , 9 A $ * 0 ! < $ G - ! < $ G  , *   @ %   / > , $ - @   > (  8  G ( @  G 8 G    5 > [   | badalo.n\NN ke\PP bahut\JJ Upar\NST uRte-uRte\PL jab\RB pan.chhI\NN thak\VM gayA\VA, tabhI\RB achAnak\RB uske\PR nIche\NST se\PP ek\DT AwAz\NN AI\VM clouds of lot up fly-participle fly-participle when bird tired go-past, then suddenly his below from sound came When the bird got tired flying way above the clouds, suddenly it could hear a sound from below. - - - 3*| - 6| - 6| - 3| - 6| - - 6| - 2*| - 5| - - - 6| - - - - - - - - *| - - - - | - - *| - - - - - -H7. H  ( G 8 K  0  > 9 H  ? * B  > 8 G * 9 2 G 9 @ / G  > .  $ M .  0 2 B   > , ( 9 @  $ K , > & . G  8 . / ( 9 @  . ? 2 * >   > | mE.nne\PR soch\VM rakhA\VA hE\VA ki\CNJ pUjA\NN se\PP pahale\NST hI\RP ye\DT kAm\NN khatm\NN kar\VM lU.ngA\VA, [nahI.n\RP to\RP]MWE bAd\NST me.n\PP samay\NN nahIrs$Ifgdl m$$Ifgdgl m$S$If^`gdgl m$S$Ifgdgl  "48@BJLVX`bjlxz䬙yhgCJPJ_HEaJnHtHhwCpCJPJ_HEaJnHtH$h)>hwCpCJPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH h)>hgCJPJaJnHtH'h)>hgCJPJ_HEaJnHtH6h)>hgCJOJPJQJ^J_HEaJnHo(tH/ #$&)*14>AHKOPRV\_dgloux}rsdfhͨ홉rh(6hgPJnHtHhgPJnHtHh)>hgCJaJnHo(tHh)>hgCJaJnHtHhgCJPJ_HEaJnHtH(h)>hgCJNHPJ_HEaJnHtHhCJPJ_HEaJnHtHhwCpCJPJ_HEaJnHtH$h)>hgCJPJ_HEaJnHtH+f}xxgQS$$Ifa$gdgl S$a$gdgl Sgdkdf$$IfTl0$&" t0$644 laytgTfhpnnS$Ifgdgl ;0b~kd$$IfTl$$  t 0$644 lap ytgThnp .046:<FHTV^`hjnpz|ݹzhCJ_HEaJh)>hgCJ_HEaJh)>hCJ_HEaJh)>hgCJ^J_HEaJh)>hgCJaJ&h)>hgCJOJQJ^JaJo(h)>hgCJaJnHo(tH#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtH0vcP222S$If^`gdgl m$S$Ifgdgl kd$$IfTl0$<"  t0$644 lapytgT !$'*-07:;<ADEMOPUX[^_`bcfinruQպ h)>hgCJPJaJnHtHhgCJNH_HEaJh)>hgCJNH_HEaJhgCJ_HEaJ!hCJH*PJ_HEaJnHtH'hhCJH*PJ_HEaJnHtHh)>hgCJ_HEaJhCJ_HEaJ7PQIkd$$IfTl0$<" t0$644 laytgTS$If^`gdgl m$S$Ifgdgl m$ "$*,02:<@PV\lrx~пৗveveveveveveve h)>hgCJ_HEaJnHtHhCJ_HEaJnHtH$h)>hgCJ^J_HEaJnHtHh)>hgCJaJnHtH.h)>hgCJOJQJ^JaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tH#h)>hgCJPJaJnHo(tHhgCJPJaJnHtH"@FHp$Ifgdgl m$S$Ifgdgl  FH &,:ƶygygygygygygygygygygygYYhCJPJaJnHtH#h)>hgCJPJaJnHtH2h)>hgCJOJPJQJ^JaJnHo(tH#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tHhgCJaJnHtHh)>hgCJaJnHtH h)>hgCJ_HEaJnHtHhgCJ_HEaJnHtH"J}jLLLLLS$If^`gdgl m$S$Ifgdgl kdh$$IfTl0$<" t0$644 laytgT:@FLX^jp|J$ n p r t x z    ðӏgTg$h)>hgCJ^J_HEaJnHtH.h)>hgCJOJQJ^JaJnHo(tHh)>hgCJaJnHo(tH#h)>hgCJPJaJnHo(tHh)>hgCJaJnHtH$h)>hgCJPJ_HEaJnHtHhgCJPJ_HEaJnHtHhgCJPJaJnHtH h)>hgCJPJaJnHtHhCJPJaJnHtH$ p r t z d!K8S$Ifgdgl kd$$IfTl0$<" t0$644 laytgTS$If^`gdgl m$$Ifgdgl m$                   !!!!!"!$!(!*!.!0!@!B!F!H!R!T!X!Z!^!`!d!v!|!!!!!!!!!!!!!""" """""""%"ijhCJ_HEaJnHtH h)>hgCJ_HEaJnHtH$h)>hgCJ^J_HEaJnHtH.h)>hgCJOJQJ^JaJnHo(tHh)>hgCJaJnHtHA%"("0"4"5"7";">"A"D"H"K"T"W"Z"]"^"_"f"i"l"p"s"v"y"Y#Z#_#a######$ͼ򮞮~l\~lh)>hgCJaJnHo(tH#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHtHhgCJNH_HEaJnHtHhgCJ_HEaJnHtH!hCJH*PJ_HEaJnHtH'hhCJH*PJ_HEaJnHtH h)>hgCJ_HEaJnHtHhCJ_HEaJnHtH"d!z""`#a####Kkdؗ$$IfTl0$<" t0$644 laytgTS$If^`gdFl m$$Ifgdgl m$##H$$D%%%%%%$Ifgdgl m$S$Ifgdgl  $$$$$$$ $"$,$.$2$4$8$:$B$D$H$T$Z$f$l$r$x$$$$$$$$$$$$%%%%Ű}qhgCJaJnHtHh)>hgCJaJnHtH"hgCJ_HEaJmH nHsH tH"hQCJ_HEaJmH nHsH tH(h)>hgCJ_HEaJmH nHsH tH$h)>hgCJ^J_HEaJnHtHh)>hgCJaJnHtH.h)>hgCJOJQJ^JaJnHo(tH&%%&&'Z((}jVVV8S$If^`gdgl m$$Ifgdgl m$S$Ifgdgl kdt$$IfTl0$<" t0$644 laytgT%%&&&&&& &"&(&*&4&6&@&B&F&H&P&R&V&X&^&b&h&j&t&v&~&&&&&&&&&&&&&&&&&̴rrhQCJ_HEaJnHtH h)>hgCJ_HEaJnHtH$h)>hgCJ^J_HEaJnHtHh)>hgCJaJnHtH.h)>hgCJOJQJ^JaJnHo(tH#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tH,&&&&&' '''*'0':'@'J'P'^'d'l'n't'z''''''''''''''[((($)&)()*n#h)>hgCJPJaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tHh)>hgCJaJnHtHhgCJPJ_HEaJnHtHhgCJ_HEaJnHtH$h)>hgCJNH_HEaJnHtHhQCJ_HEaJnHtH h)>hgCJ_HEaJnHtH((((%)&)))*``iVS$Ifgdgl kdH$$IfTl0$<" t0$644 laytgT$Ifgdgl m$* * ***** *"*&*(*0*2*6*8*@*B*F*H*L*N*T*V*^*`*d*f*p*t*|*~************************+ +++"+(+6+>+D+J+P+V+ŴhQCJ_HEaJnHtH h)>hgCJ_HEaJnHtH$h)>hgCJ^J_HEaJnHtHh)>hgCJaJnHtH.h)>hgCJOJQJ^JaJnHo(tHAV+^+v+~+++++++++++++++++,``` ` ``````gaiabb`g.h)>hgCJOJQJ^JaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tHh)>hgCJaJnHtHhgCJPJ_HEaJnHtHhgCJ_HEaJnHtHU#hQhQCJH*_HEaJnHtHhQCJ_HEaJnHtH h)>hgCJ_HEaJnHtH".n\RP mil\VM pAegA\VA I think have is that prayer from before only this work finish do take-fut no then later in time not get get I have decided to finish this work before the prayer because I may not find time later. - 5| - - - 3| - 3| - - - - 6| - - 5*| - - - 6| - - 6| - - 6*| - - - - - *| - - *| - *| - | - - - - *| - - - - - | - - *| - - | - *| - *| - -H8 , 0 8 A ( $ G 9 @ . H   0 8 G $ A  0 $ - >  > | khabar\NN sunte\PL hI\RP mE.n\PR ghar\NN se\PP turan.t\RB bhAgA\VM news hear-participle only I house from immediately run-past After hearing the news, I ran from home immediately. - - - 6| - - - 2| - - - | - - | - *| - - *| - - H9 ( M 9 K  ( G & C "  ? ( M $ A , 9 A $ . C & A 2 8 M 5 0 K  . G  8 ,  K * M 0 G 0 ? $  0 ( G 5 > 2 > - > 7 # & ? / > | un.ho.nne\PR driRh\JJ kin.tu\CNJ bahut\JJ mridul\JJ svaro.n\NN me.n\PP sabko\PR [prerit\JJ karne\VN]CV vAlA\PR [bhASaN\NN diyA\VM]CV he firm but very soft voice in everybody-acc motivate do one speech give-past In a firm but very soft voice he delivered a speech that could motivate everybody. - 6| - 6| - 3| - 3| - 3| - - 6| - 5*| - - - 6| - - - - - - - - - | - - - - - - Table 2 List of POS Tags and Multi-word expression (*) tags used in Table 1 (a) and (b) TagInterpretationTagInterpretationTagInterpretationNNCommon NounVMMain VerbPLParticipleNPProper NounVAAuxiliary VerbCNJConjunctionJJAdjectiveVNVerbal NounNE*Named EntityPRPronounRP```!ahaialaK8S$Ifgdgl kd$$IfTl0$<" t0$644 laytgT$Ifgdgl m$S$If^`gdgl m$bbbbbb b"b&b(b,b.b8b:bBbDbHbTbZbfblbrbxbbbbbbbbbbbbccddxlhgCJaJnHtHh)>hgCJaJnHtH"hgCJ_HEaJmH nHsH tH"hQCJ_HEaJmH nHsH tH(h)>hgCJ_HEaJmH nHsH tH,h)>hgCJOJQJ^J_HEaJnHtH.h)>hgCJOJQJ^JaJnHo(tHh)>hgCJaJnHtH%laHbbFccccdddikd$$IfTl0$<" t0$644 laytgT$Ifgdgl m$ ddd,d.d4d6dBdDdLdNdXdZdfdhdndpdxdzddddddddddddddddddeeee,e2ehgCJ_HEaJnHtH,h)>hgCJOJQJ^J_HEaJnHtH#h)>hgCJPJaJnHtH2h)>hgCJOJPJQJ^JaJnHo(tH h)>hgCJPJaJnHtHh)>hgCJaJnHo(tH0ddde.fffff$Ifgdl m$$Ifgdgl m$S$If^`gdgl m$S$Ifgdgl dejeve|e~eeeeeeeeeeeeeeffffff}l]MChgPJnHtHh)>hgCJaJnHo(tHh)>hgCJaJnHtH h)>hgCJPJaJnHtHhgCJ_HEaJnHtH#hQhQCJH*_HEaJnHtH$h)>hgCJNH_HEaJnHtH h)>hCJ_HEaJnHtH#hQhCJH*_HEaJnHtHhCJ_HEaJnHtH h)>hgCJ_HEaJnHtHhQCJ_HEaJnHtHffff,g0g?gCgRg}xskMMMMS$$If^`a$gdIl KS$a$gdqSgdqSgdkdT$$IfTl0$<" t0$644 laytgTf,geggqs{|OPQSTVWYZ\]^_`abyz|}ٻxqfhphp6H*] hg'6] h 96]hphp6]h 9jhTUhThjh0JU h3:9NHh3:9jh3:90JUhIhNHPJnHtHhIhQPJnHtHUhIhPJnHtHhIh5PJ\nHtHhqPJnHtH(RgVgegS$$If^`a$gdIl Kegfgig2S$$If^`a$gdIl Kkd($$IfTlMֈw F#;/V t0644 laytITigugxggggS$$If^`a$gdIl Kggg2S$$If^`a$gdIl kdX$$IfTlMֈw F#;/V t0644 laytITggggggS$$If^`a$gdIl ggg2S$$If^`a$gdIl kd$$IfTlWֈw F#;/V t0644 laytITggggggS$$If^`a$gdIl ggg2S$$If^`a$gdIl kd$$IfTlMֈw F#;/V t0644 laytITggg  S$$If^`a$gdIl ParticleCN*Compound NounPPPost PositionNSTSpatio-temporal nounCV*Compound VerbRBAdverbDTDeterminer/ DemonstrativesMWE*Other multiword expressions  This work was done during the authors internship at Microsoft Research India.  Verb Group here means a group of words with the main verb of the phrase as its head.     Proceedings of ICON-2009: 7th International Conference on Natural Language Processing Macmillan Publishers, India. Also accessible from http://ltrc.iiit.ac.in/proceedings/ICON-2009 2S$$If^`a$gdIl kd$$IfTlWֈw F#;/V t0644 laytIT-1FJXS$$If^`a$gdIl XY\2S$$If^`a$gdIl kd$$IfTlWֈw F#;/V t0644 laytIT\cfsS$$If^`a$gdIl !S$$If^`a$gdIl m$S$$If^`a$gdIl 2*(S$a$gdqkdH$$IfTlWֈw F#;/V t0644 laytITOPRSUVXY[\]^_`abF$a$gdo4aFEgdӍڍ hqPJnHtHhTh 9hg'hphphp6NH]hphp6] S$a$gdq?P&P:po4a/ =!"#$%0 DpC 0P&P:pw/ =!"#$% P DpC 0P&P:pg/ =!"#$% P Dp? 0P&P:pg/ =!"#$% DpC 0P&P:pg/ =!"#$% P Dp? 0P&P:pg/ =!"#$% Dp? 0P&P:pq/ =!"#$% DpH$$If!vh5##v#:V F t65#ytCTH$$If!vh5##v#:V F t65#ytCTH$$If!vh5##v#:V F t65#ytCTH$$If!vh5##v#:V F t65#ytCT^$$If!vh55I#v#vI:V F t655IytCT^$$If!vh55I#v#vI:V F t655IytCT^$$If!vh5@5#v@#v:V F t65@5ytCTDd $%0  # A"sۣ~}i ?O @=Gۣ~}i ?`x[3x?fI,7,J,,X!&)hM,DXXVnAAEF@b+#B5Y{̽s߿w}.1O{;:v Y.\p򯎳g-vWK.oܸѝ;wv<@cok~w֭ؿ˲ݼy6s]ڶ,}G2Μ9ï^Z=]wQ;rʕ+ĉ~9jlljGM͛7;^իh[nuׯ__=g}LOVŋ'\reҾ>p[m]l==;r qƪ5^,~mw;C_#3M j3g`3-,[1tSVGm2k ϝ;(.\@j+txBwcI`w>}HdҳȏzV#A?^l]tMdrDmfycs+ic}fuȶ 5NmsbK_l|}/n.`ҡߪkeUqry6*CZ6=C=չޱt=~z ~]C}Wnskӡ6>9OϢ{XF-%6E}>sT?'-uHmǶX%ztGNܘ¦ӛEwlZ~j?j'Fjo=|%Zo|oI HsϜ>ƾ:y:{m}CҟSloG.l\*6d6>W+q-|Dž!ْM4}Oo}q&d{[csj՜uTv}=1.ʒ;}fˡ|lo5׶cڪFƌɥlyArsTlfKQVcZ h"}m|tmϧlk*Jkjb_JQ#-vukbS֙G7l%-t,&2;KeK%oN65%c[u򓓭Rl=ϒoi|^CKGm5Rc[R]wLNmʐ,5 No|n1ȱ8gzt^|ߧrQ;״/=q߯ok7jIgm.s>Ryӱ|]?P]/7GM>8Sshuպ.u-27_Mo 52Tn;׎1|ccֲ`\lQ_d-}|[WtA{^?i6xr:ٖz\Ϋ\++7}>sԯwc"}7ioi/9tn#>c;H>V*wtܮ=/gu]iJEtPۖFQ>jen̘AΏg7otzON 1Cs\蒏-E+$ؽ-/Li}6d_߾{Ӧ3K9<>wƹHίi#C/t&ʴsTZu\Y3붿Nϛyjγ,b\.' ~EԚGG,[̍|!>G<6gnڲuQ|\&ߠ$9˅M?>VfVon9n>^_nӕ >w̦ٶq\Dz=k9ubAt}(rmq%]SW[i(]8v]ca9rՒn[|ߚS:Զ%ݙ[\+scXq[Gotuz3=n=J>_>ܚ6&3OW;~{ܜR9-m;TWޔ̩uHmV FnZ3VPR?MMæX/={rp[y\Ӟ8kY5_ztyl/Ȋ'X^Oܫ|m<̦_6eʶ.>~O.ru N8^W-t[s>-rwS/:Gm^گٝcγgme-oM6m+fҥ527&\9[4N.gMæ秋);JC{LΩ}>Bмf[ۦ!ys<5Ǯc*"ɥhcռn7c.ZOK!y Y[uhƮSˇEw+[wM.O] s7MrkUUrZsCݶtӛɗyh5q56wP#ߛ˖'mT"״tӔbqZ7.ߵZ!Sh[lM=6g9kx5^r<pi(-5^`9q "7MSoKOݥ=mw-uȡvcޜ:)wpٳ{jy[]0vO. @~WP }]cuoFS'{w|I:ؘ?>,kG}A}/.X_5unGytq_L>?7|ӝ>}{g>G>3 w^[Th{n[_x&4/_^ Nz=* `C?Ӿ5.*Y˖M{7rw饗^z|*)}חa]rV]٩bU6G}tre|zIkdOڮ"`{Cu^tSbU9k'x ZOf]sWJyM9h?Ϊ\T,3Osu| =|`^v}us08k:_z S9ty"suWҡC`A qxLi]D*su?~;tSO=ս۫*ݨN^Y-=bĥ矧+tּa,î+1NҿS?*~%05?J{0:ud_obsCS`9vLr>Uy?j7G\Z}ZY1dk&꫓'F+b-˳ҕҟ-EOȆ\%Lݮ{駛]1O\ӹ"SKb,AX)kٶsL:t٬Xl?x«Z{-ݶ1Ӵ;6e="4gu;8lk"6GNysϷS[g)@,`~t'9u;[Х*z^ZlyeqN̯km]Ķl; 2싙_t@}ΩEW8Y喾2,*SZNhQ7^sݴ_;a=*]բ}rG['@{d㳎;1v'еu/3OĎPYoIF= h:sPvL+@9ӭ!?*tQs69t~f4S~8E.9Zzs.UQGb,#؎7?"u ͢MڄgP. /,W9Snڎ湾/: `9]`=mCOEs%{+7#(#`HX'`Ӈd5;}BnŲ$|ִw_`9?@zi>;6㬲X{ɾ#`yZ2g;ə[»I`;{ (1`⭬g18\5{TsF`u~FVORUVLp<GO4?} s=]k/0\!Ъ."Y1b1?AblnEyWlUV`>;|0'fKbSi0B qqw:؇4~CHF5爇PiM`O~x`_8TS_ey8kdz|<ؕɦ `7c+ߊ(+H&Ip:>ٟ8xe$Rc&1}i9;O㸾F/}p/WXmq|'[lw!`>hl۳ݏ-n-c %hɮ vo{S[خƏ In|H? hբ6TO~w.C5E&W:cwm,#Ϸ?6Rm>8sy9#m Ylx]X>7O?rDd6EFF  C "A fig3R~q?iZq FRq?iJFIF``C    $.' ",#(7),01444'9=82<.342C  2!!22222222222222222222222222222222222222222222222222d" }!1AQa"q2#BR$3br %&'()*456789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz w!1AQaq"2B #3Rbr $4%&'()*56789:CDEFGHIJSTUVWXYZcdefghijstuvwxyz ?( ( ( ( ( ( ( ( NEp. PɶMzako&\cEG[qE kNT!T,#IO_t"8/4F-A-%ps?&'=g֞:߃o &W<ӰpvUQs`s@WuPpyÖgm':U?ڇ=lgF2Y!*Δt? i:S0g`1@'\Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@/ xWcmGHN3Sއ>Y1oc n\,Ik~[qHlm񙶅$r~?+]F<᷏<1[ᶚO;2I>6AQ}V حb[ƙ%ۖ M) M3k@#ӡ(@Ebq.x5·žʭ̚n$'=>uh6zUQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQE|[d`ҭțǚpaw^k//OΩwf&PPܓߐA{+uXX^UF+ⷁտv;Ca}zW? vp0FG]eCQcnr"0Qwsm> mfj6I(6q-޻->KM`ēF` bx8ῈQtvMQ8$6I?/AQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEVG;i}&9J mxHee="ƥ"Eg.,zzuǚ(HDW-#\է/4^Ct.cdBx((((((((((((((((/Ɖ&YKOB0 ԐkoDWĞ!8u~2~@ 2j4=?új\>EE=I<@rLݠLBIc'8aϭu^rW!vǢ=: 'OͶo 'z=ghdt&8rYG9f$I5@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@y;Y:v o#Z숹fEl0AϠ5 好wQ, R0A(/:%|+u>mod^J\} rGX-SGu?"V l'aow^mvmƱDŊ'[k8{а +B=NBDFŒ;pǫmP?&j}&Y4 cH=+F ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( FeE,$VP|#"Ǣsf]>KjNmRsǠJ=XOO}#A}x^"iV9T3yrlvvQ\2}3Yux_uj:ȊC)"Yg-~ݤ+Z4a4(-nF)t8xaASԵbp(((((((((((((((((((((((((((((((((((((((((((((((((((((&HXF.%Srqgݻa뎴lMط=6gcuf8>Ku7au ..77#§5vVorkX-""z>I55Tr(( VKp29dDZ4/Yf[kXa%tyjG<\LѺTF&J+;̟Lrc^!]}{"0darՁ;) (((((((((((((Џ;w#"!}OҸidGgUu, N3ϧYrYfWr զ^d~G4&%{k*FQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQUnD2!O:V q?=,i\Mؒ+HǓT @;TVQD1nB;g/.κ#;5nVl[ 0#O}jscKw.ǘCF?Q؊Kk?"NUǪw˔QEHŠ((((((((((((() OP{,|%߄sJ]=BzH>S/ҬZj $=Uܛbo =-oa5~85$TFT`)RnJERQEQEQEQEV{h6i9{f8F=uCִ(-nF)t8xaAS[%ur 2uЎd-,W0}Ji\M.ndy~hG>J;Xf9,{{-ˌfcǹ'ho) ({jnbSL}]BXT;%QQA5b^F (l=>Mf Yk^'ie(aEPEPEPEPEPEPEPEPEɦW2IwHi$`,pT;R]V&3{NHdgBw=nz_}I*+x FQĎ 8?u;Uʥ ? ~QYh`ImѺSTWbJ(((((((uzP+K;>{s4 ؖiⶅ#^j>YtyG_*HlIIE}=OGLt﹛2 .cul p8`{s# spmf=2|K64h((((((+7Q%tVqH"=GN}i6݉ndy~hG>J;Xf9,{{m0lލr]XQE# ( ( (3%{:Y?1x1L!xePѸ{e47,Zh?=~z;h^*J ( ( ( ( ( ( ( ( ( ( ( ( +"Y/M4n$Ra.ʌʜm zw.]7mb#^@Q@Q@Css^dpY`sL[meOq/V m,ia%}T3@R]XEm-̫sz*s9Xϫ~E%`) *Ռ7{Y$%tfi`jwl8_25=vuу# SE:O{witXX,?t'~ ѓ4(vK(u=}9kqp) ((((((0ԍFY󎧊,%`r}*6LWSD~ս.WeZ||9X=O8WQE%Q@g/Kt|r~;wQq=ǡDU; =q~돯v5g`NERQEQEQEQEQMwXѝ*(%=j?^ݒQWߩ펥q7a^I4mbS%ad~$Tֶqچ`ZI%~Yހp*Xa%TFT` }6 K (aQ\[un F8 Ah vW{K C;*qGcܪ:Ku?잇>n(tnè?瞵O]PE# ( ( ( ( (WM1lH?ˌcv{ҸK18ǰOloo"(rI9$$'֋{xaDQ''>- @Vg0ʐ?\{uh#"20k9OT3.px'ӧLa^Ѣ*J ( ( ( {jnbSL}M;W+ K1ʇd0?=U aegG2'huc&%{k}(aEPEPEPEPEPEPEPEPEPEPEPQִ#KDK8x993{W1Q{vOswt&oIl, '}ʮYC+ FA?>zƱy#[٭(Qle>!j׎muLcky~L!X+ wpyZjͧ<, 33m#v0Oj+<mYDž.|;YZ\E$i.$ciaw5QEUKD$d?u? [Zȁ[2#_VCb+X9f',ǹ'&mnvc,XMܤQE ( ( ("#LA}SL{j?~VӶhDH!#>ڥwz)F8of^?Q؊bj :bcNbsv½/E%Q@Q@Q@WfsaywV3'';]I 2RR9W ?%qtWyCI\>)x7zN}\_e7mRǖ@MvQEQEQER@Rw\AD?y?ެ4w0$6C)+9]}#FϭRX(((]?a+7@Ey.߇0oJQ^ ?%qt*A !ڈ8XG4-om =B? d_p<Ws~Wn\}r}c:7: %H=\d Vwڅ7s=9cl)Ao|o ^)kko0ZRJ€ (o[Ӽ9ϫjgo.}(A'/hTd8V\%![U]-<ֹ6锫x#+y{Zu:5YCrZJ>(9G@+tDWF 2)k~'+_m}W]k2KYU=Wc@Q@Q@Q@U)yѶ[62"Տn-+M3ZZ8[62"սu=oo"( 9ܓܟZX %ڣ$Ԕ7]XQEEC# 2G:ωOmbqrTq=uL(煢 TKYQ3h ԝQEIAEPEPEPY˽,Bz}'շ{81.XG'5U[Tcԥҵm6k{ȣW,ʷqlx-ߥp~}}OfxC\[٤C|mf;nṃ8쾕*Ӿ;CVռ5$*v& y\Ey'ۂNφP|A ^k3 mf'6+bv=8Wu৆#MK\d$Pǽz h6ީŨ2ZY3rkO<5`_PEPEPEP;fcklJlcu=BDP{ܓԔSQHaEPEq~+O\z^Zڧ@="xwxQCwRFu`p$^mBvCmsX9! *jsӛd,HK-?9IbpʰE u@F>(0ڎii} =卺05f|kiirj,k"`zG(>w.;'fq uQڮi:l:6c۴v*p=\9xJ)- ku[:2ޣ$&|g ]UyHe[5 ;j1{FcB䂊LP?4w߈n~!x,֭<;h:ygV*G=q$3ZMk0Ynm<{BJ{Py_?d_C/Z"~ck7\vŠ(z_j #Aeo%ċ dQW(>|ODŽU^tVd priݸ29]FᯉXu C*\C'C?IV׋~ v+ x;CVbE`§8e?x}<y7xbM9skOgy'H#1]4w߈r:sCšR%G̀VyG5OƟ"-JX3Nr9=; ??vh7\c֑MO mL$\ s\/4w߈n~!Gx?a߉~|D%Nӭ 3*0ssڳv<$?sn~!NX"{ok S^GFS[9<[A&KGZ(((_]gp/Rǧƍ2hcCFk)*G+XܱiT=Ɦz((((((((((((],žmˌy_&W:mF$qKAz e ~tK01c_zZy mˏR1GYݴBaETl{V},mnŠ>)Tae 2g7B{Ү[kmw[_A-ߴ@ Eڐ˴QE@ʹGlX[]{rC@b>N) ͭ8{zS`U(ĕ(0(A&k9y11{~A4ӰҊ΋%, rczV XQEQEQEQEQEQEQEQEQEQEx*My_hvZܴ12]HܸqB7'qey+jY|fnW# a{v(';Y [kުGJWÝi?Bܼ,KH({@Ny~=7u2i:Ξ-beiy?U6O\|D]IOm7QИF9U f;8._)#\V6Zƣs?ؑn=ޘqGʡO@6jڎSՐŸi p7( i TԴQE!nc۟x3[Je ĚޡU 6$c0'=/v}sOƕȲo٣G]m^tNϺu*s QzlP|Ku?mr$NA`8PZ7__֕{&Kw`C/ Ppid |?YIP =0gWrqE+m8?-sI5D֗(c>xQaawPyFN MQZqơUKppbg^9t_^`˳+̅!Iv=p+٨(((׶%1xoPzj98b #$z/EG[o'"?و'XjkW-5WJcXnVGPsAV,Z[Xr.0}GPy\KGoR{Ѣ[jP\KkGf#ޭiƝ (Š(((((>%44-D뗢2 \ىa57[ۏML. **[*+wQry~gC}asͬ˺9bl P[Zlg^G9g>TQp ( ( (+ڋksFwE *wwFY$_.#XzPEY6-c toc '} ]$@AFᑇP}j0(('O GQ\~u$BeO&P)TU ?%qԼ57 qhmKTds^CcCF&I5cr!qe ?%qu{3xk@2K#Ix·^~nK54xqwp}tyCI\>/^0:ڻ;vbLtxbWLT 9Oxᗋlu1Ͱ|9ϕ='/! Na2mdPjA?  tkWwRy9 MbKI(0N Y e &dU[{s 9 dg4C/$Q ?4?Y-jZYxWEuWGƮ2&`9<}i&}N[K'Zw.p9]ݺϨ7$?LEc|JiO O 5{[w Ook'ƽr~ѣ5Mʗ&}=RGSze,Kmt[sN b: =W+Ѧe X`bڸc~w{=zJ((KsJq"[K%g>D-G>PGuLYN Ѓ؊ibk&Z\]np.c/M'p) (((((((((((((((+ADu˩9eA-2=rrxڽ Z/|`Ku K\Kw!h+AU#!pr1ϷEQEU)8b #$z/EG[o'"?و'XjzyOnGXa#*?y?A??W !IchK`"XV$Q(nm`˸d\n=E6,gl8uE5&EKmJ |2`md{պ ȼEF_sm/vÏ_ѿN+ETԠ;& =G[[;QE!Q@Q@Q@_kƏ=;RE_iTLp"I<p~ g^JT{)C4k׍HFrk|w$Ŀ M^sm/QֿfӴ"{{\erhq{Qƕi-B[p1msmc τ xVT.(gmdERw#JGH $tQī&$8-ϧ8߄9Mzo%* .䃑v>T8?ZQVltidiM&晔?}OڼIӼUx:}KZ-rxu8^kP$`98z3%>UsI]Y- e5hF `#W7mk L-Շݵ<2I䞔|YM¶wM,^mouOjZz-}ay((2[omcD?媎ק;RXiԝ4h9%Q@ Chiu;PNm-zg،<l~,jxsOeBzCW!'0Jk*2jVț󏔞ps@}'J퉷As2m8$@y1>O ZMC-}x}[[kp ro\e7po)̗#F}/4mVZfy" bi>oxoZowTI\ΥQ;a0Rz 褍%pUC@>,|'iӢE}HzzP'Tۊ%D/4emA-ڱ}xSxoZowTI\ΥQ;a0Rz (+O^(l^ RsC$eč$_U;'MO>LB$y뵈(4ÓiR7wmrME13Vg'dQWW7iѦ[fuf UWjv(((yV?-{o i.Nډ\1VFHyGxo = g|1A<5ZS:+{u!=BG#{/xڶm!KtCc.A>ؚWYi'QKGx̑v s!n:Q-젊o'D072MKA}"K/ ;W;TV'|E<)kG[<* ۠(( viEd.xvPzIguet8ZFUu*HdYgӹkQ؟?''#?64hm̅.pA*{ (((((((() NMGss^d͵sc9'U1m6Cޮ~kCޞDEp\Ah%3O,3M y[#dR{G>P]ڭ2t#Ust.^I誶WM:SKNTtaժ4QE!Q@Q@Q@5dFGPXdN8럅>K<5d!?R";+]:+K+x]U (()M-,Ωc*Զ+Vwoaxy&[Y&S?IuboSKk3<~=n٠7ټ|n=1RIHs{A,̑I]+DHqIT>qAr~I=wN)=f.U-x@d;[[;QHaEPEPEPEPEPEPEPEPEPEPEPEPEPEPE5cFw`bp5>"/ofzdi\MؒkyVYׇv{G?AIkdҳ40?S=u,0Em *FEIMŠ(QEUl^)Z)XorGcsWʶpCp/ zqAug (r)!Uk&ѓQ>?>)={nhQE%Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Vfp8U?"}Khm{[;?Jiubow5X==U[H#+,ԞU T`8}%ՅQR0(͈_[]0 ?~[0ȹ?tg+'Nӽ\-Ẅ:Cc؃'ъ݉h2Mr6fH@}9W%dцUi5`N㨢C ( ( ( ( sz!A yL2"{Y[hl茪AҬZEiXr}Ijmɽ"+/n\Mu1胰{(ƕ(28H{{kh݈v EKY zOQ'}:n{sK] zjh@ A=(((((((*HXfSIe#>:w= hUyWnGC*"*UhՐjgW9g>jE%`)QE ͬ7qys&T>}};&[n rqhM>MuX2R AZk9lɧڱŸt?}n}IeRFpz ERQEQEQEQEQEQEQEQEQEQEQEQEp j76uk+2}sԐ!0pA@~Wn]?a+7@Ey.߇0oJQ^ ?%qtW?xw?j>]wdn3Һ (({5 w˺>9Ha8 ]?a+7G.߇0@oJ ?%q( Fv%yX0OHÀz0b ( ( ( ( ( ( ( ( ("kh^.7JjIREQEQEQEQE6۴ 6\tZ((((*9KZ)A(0 F*DEB 4gv 2X)jZM֙|n1Jt `PAGzZ񗲾-爵]"t]r0 Anj^ރʅ#ϱBs8O>(20hZii0^t T!% *8JOQ4Wմ땺MV :^((((kHau <۫I-ԁXK2B>5iut8mby088h?D/jOt;yrٻk<4A _H`-xNC·wk|Fpz0c&h"=oǿ9ϫj`K~py p+?'kڞswJ^s]~ƌd9'oUڀ=B((kHau <۫I-ԁXK2B>+ֽgt9|_)i%|ZEΆFh-N;GnҶ?!Bf((>%|Zu.S·֩/f-Tc+9^]炵_xnvh嵐3w.WicӰ(((((((((((,?7ӵ]n+G,OW9fϩ'5uHt/^*{_@%6`&A嚀8Z;k8 |+-JѾ! mId:ux%}mm#29t$v/Rpּ?KL_#ZE%16+-ۙI`OÑo[#O/skn5ka6Wd#G k'O hSJw}>ѧGyZߕ95 [FaUNX}wu$ե-c&\ȉaےB K6ڍɊ+h /x&^^ u0 +8 Z >C^)5JY=ԪC cay^9O|?j~I7m }1ziy:U3gqօPEP\?'C|K?}>?3ٷwgwKާQҡLLPRha[p'G#4| ?5a oZieD_)#9p{|9WV:D,4E 0˺5]x=cOk,ok;"F,B8]4!d,Fi!Q;sׯwZno03haX r((((((((񯂴hzp+Z ˃Xc{zWOն/$)5?Íww~vsv=^/sjvf ^iӐa~qWys@|@W?Kq<++ J7_j塳c22rW?:޽}[JQ縻E|=Gr(3%p h><֩]P۪ȊB#dF9˞ s8~,&|[]SKԦ[ԷeBvڊskoȥS@&x@W gJc[Y|a 7P,qZ qq0qu 3p71;clg$ؕU#xcv]`xZ@|Ŗ'ˀt/ ~>O׊>۪ʵ<6?/Ɍyy۞;moO{I{O5r\,O^=oI5 <8b=zb6*ì]H^o#F@`Ig(珈<7jZ%կ'2t FF=O_+WH.5 egEht @d'>"R9uo^^WY7unNe/z}(XDTQ9 :_^=VK98Ue>5kZ{*H)YNSu}?#=ˢƶ1jPH '-@5 Z&;!/)班v_>Uj Ԏd' ?vA{!w{ՕCE'Gx ?_eo`4{FͰ1d*?y' ӎhAmuM.RyoRݖDe Q#j)Pwk/Ínow}ѱv1W{WG<+ (9+MOw \-µ`;0緥xFAˡ>+HMq,"yNck+O[H|-b܈A !W83_&}1[[vpp\rYAmuM.RyoRݖDe Q#j)Pw_WmzڦwM2!TW֓-͕c&Wq1ïIljYMh&`!h%$B(ɮuǬ?o# %XGev.NN zo/$:DN߈u:7O J-bETwSj7h'ѼCo5\1D <ҡOůϠh^>_^1/m5) ֯EŲE+aBz~ռEh+ ڞv.ܩy9EG3Kֵ}y X.06e/R3~i~5|yDxgTໆeEV,#lqֻ'# Ckۤ@)W'#?t>T jG޸&mHu-(4R1Mdps; Zܾ$ukǗz׉vl̠VeRe!Ox~/&9Q<<& UP,r9<9/S\)z 4EPEPEPEPEPEPEPEPEPEPEPEP7_i<ѡI+VuV,r%ЭO눮1v7U`=C ժhwi<yr O'[?-Q;Ɵgm-Ac$(9eai Y *W''I:g\/<^;94sLj,G'~:QEQEQEQEQEQEQEW( <՛öFT'9ϗ|f(5 ֏oދd.[ $^5xs?ڠ=ZX, `׮m5{ Zh^Mm=M.p );>uᯉVk6Ofl,v|?O5?Ud(=<­Zw+Sky) Mœd('@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@Q@fk sUN1#85Eyw3.|C>xWXXZivok(EY ( ( ( ( ( ( ( ( ( ( ( ( ( ( ( =K2𗎮`[ciƳ"N ^zm(4ÝvOY!]^C  9K((((hfbZắ>3Gi4MVRn^MKc+2u ^#E]cI(F YA?T? xZ_;FmcI=@w$GEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEQEDyK http://www.praat.org/yK Dhttp://www.praat.org/yX;H,]ą'c$$If!vh5$#v$:V l  t 0$65$/ p ytgT$$If!vh5&5"#v&#v":V l  t0$65&5"/ pytgT$$If!vh5&5"#v&#v":V l t0$65&5"/ / / /  ytgT$$If!vh5&5"#v&#v":V l t0$65&5"ytgT$$If!vh5&5"#v&#v":V l t0$65&5"/ / / /  ytgT$$If!vh5&5"#v&#v":V l t0$65&5"ytgT$$If!vh5&5"#v&#v":V l t0$65&5"/ / / /  ytgT$$If!vh5&5"#v&#v":V l t0$65&5"ytgT$$If!vh5&5"#v&#v":V l t0$65&5"/ / / /  ytgT$$If!vh5&5"#v&#v":V l t0$65&5"ytgT$$If!vh5&5"#v&#v":V l t0$65&5"/ / / /  ytgT$$If!vh5&5"#v&#v":V l t0$65&5"ytgT$$If!vh5$#v$:V l  t 0$65$/ p ytgT$$If!vh5<5"#v<#v":V l  t0$65<5"/ pytgT$$If!vh5<5"#v<#v":V l t0$65<5"/ / / /  ytgT$$If!vh5<5"#v<#v":V l t0$65<5"ytgT$$If!vh5<5"#v<#v":V l t0$65<5"/ / / /  ytgT$$If!vh5<5"#v<#v":V l t0$65<5"ytgT$$If!vh5<5"#v<#v":V l t0$65<5"/ / / /  ytgT$$If!vh5<5"#v<#v":V l t0$65<5"ytgT$$If!vh5<5"#v<#v":V l t0$65<5"/ / / /  ytgT$$If!vh5<5"#v<#v":V l t0$65<5"ytgT$$If!vh5<5"#v<#v":V l t0$65<5"/ / / /  ytgT.$$If!vh55;5/555V#v#v;#v/#v#v#vV:V lM t0655;5/555V/ / / / aytIT.$$If!vh55;5/555V#v#v;#v/#v#v#vV:V lM t0655;5/555V/ / / / aytIT.$$If!vh55;5/555V#v#v;#v/#v#v#vV:V lW t0655;5/555V/ / / / aytIT.$$If!vh55;5/555V#v#v;#v/#v#v#vV:V lM t0655;5/555V/ / / / aytIT.$$If!vh55;5/555V#v#v;#v/#v#v#vV:V lW t0655;5/555V/ / / / aytIT.$$If!vh55;5/555V#v#v;#v/#v#v#vV:V lW t0655;5/555V/ / / / aytIT.$$If!vh55;5/555V#v#v;#v/#v#v#vV:V lW t0655;5/555V/ / / / aytIT^T 2 0@P`p2( 0@P`p 0@P`p 0@P`p 0@P`p 0@P`p 0@P`p8XV~_HEmH @nH @sH @tH @8`8 [Normal_HmH sH tHb@rb [ Heading 1-$$ & Fm@&^`ma$5CJf@rf [ Heading 2-$$ & F2x@&^2`a$ 5CJaJTrT [ Heading 3$$x@&a$5CJ\^JaJDA`D [Default Paragraph FontZi@Z [ Table Normal :V 4 l4a _H(k ([No List NON C)FEACL Text Indent$`a$PJ :O: B[ Template TextCJ:@: [ Footnote TextCJ@&@!@ [Footnote ReferenceH*B2B )F EACL Address$a$CJPJ 6U@A6 Hyperlink >*B*phOb )FEACL Abstract Heading.$ & F ^`a$5CJPJ aJTbT )F EACL Abstract$TT]T^Ta$PJ .r. [Text$a$PJ LL )F EACL Title $,a$5CJPJ \aJD2D )F EACL Author$a$ 5CJPJ JJ )F EACL Email$<<a$ CJOJQJZZ )FEACL Reference textx^`CJHH [Template Example CJOJQJDD )F EACL Example  ] ^ ZZ )FEACL Example 1st Line!!]!^!PP )FEACL Example Last Line VV [ Template List $dh^a$ CJPJ aJNN )FEACL List Indent! dx^ T"T )FEACL List Bulleted"B^B`V!"V )FEACL List Bulleted 1st Line#RR )FEACL List Indent 1st Line$x8R8 [Address%$a$CJPJ >b> )FEACL Caption Text&HZrH Plain Text'OJQJ^JmH sH tH NN )FEACL book/journal title(6vv )F EACL Section2)$ & F T>>Q^`Qa$5CJPJ \aJ66 D)F EACL Text *`jj )FEACL References Heading+$xa$5CJPJ \aJbb )FEACL Subsubsection$, & F T>>Q^`QNN )FEACL Appendix heading - & F\\ )FEACL Acknowledgement heading . & FDD <EACL footnote text/CJFrF [Heading 0$a$5CJPJ \aJN"N [Abstract Heading 1$a$CJ\dO"d 1Abstract22$ & F 7]^`a$CJPJ R2R [ References3$x^`a$CJPJ JRJ [ ACL Title 4$,a$5CJPJ \aJ:R: [Author5$a$ 5CJPJ @b@ [Email6$<<a$ CJOJQJ:qr: [ Text Indent 7 ` :: [Example8  ] ^ @@ [Example 1st Line9xFrF [Example Last Line :xDD [ List Indent; dx^ TT [ List Bulleted < & F @^@`LL [List Bulleted 1st Line=xHH [List Indent 1st Line>x2"2 [Caption?CJ\44 [ Caption Text@jj < Table Grid7:VA0AP!P 4ATemplate Text CharCJ_HmH sH tHF"1F 4AEACL Text Indent CharPJ 42A4 *4AEACL Text Char4@R4 o4aHeader E !4 @b4 o4aFooter F !dOrd HCOLING/ACL Text Indent CharG$`a$CJll G COLING/ACL Text Indent Char CharCJ_HmH sH tHDOD COLING/ACL TextI$a$CJzOqz K#COLING/ACL Reference text Char CharJx^`CJhh J(COLING/ACL Reference text Char Char CharvOv COLING/ACL List Bulleted#L$ & FBx^B`a$CJaJJJ NCOLING/ACL Text Char CharLOqrL MCOLING/ACL Text Char N`ff COLING/ACL SectionO$$ & Fa$5CJ\aJrOr COLING/ACL References HeadingP$xa$5CJ\aJ^^ COLING/ACL SubsectionQ & FxCJaJZO"Z COLING/ACL Text IndentR$`a$CJfO!2f COLING/ACL Reference textSx^`CJPK![Content_Types].xmlj0Eжr(΢Iw},-j4 wP-t#bΙ{UTU^hd}㨫)*1P' ^W0)T9<l#$yi};~@(Hu* Dנz/0ǰ $ X3aZ,D0j~3߶b~i>3\`?/[G\!-Rk.sԻ..a濭?PK!֧6 _rels/.relsj0 }Q%v/C/}(h"O = C?hv=Ʌ%[xp{۵_Pѣ<1H0ORBdJE4b$q_6LR7`0̞O,En7Lib/SeеPK!kytheme/theme/themeManager.xml M @}w7c(EbˮCAǠҟ7՛K Y, e.|,H,lxɴIsQ}#Ր ֵ+!,^$j=GW)E+& 8PK!Ptheme/theme/theme1.xmlYOo6w toc'vuر-MniP@I}úama[إ4:lЯGRX^6؊>$ !)O^rC$y@/yH*񄴽)޵߻UDb`}"qۋJחX^)I`nEp)liV[]1M<OP6r=zgbIguSebORD۫qu gZo~ٺlAplxpT0+[}`jzAV2Fi@qv֬5\|ʜ̭NleXdsjcs7f W+Ն7`g ȘJj|h(KD- dXiJ؇(x$( :;˹! I_TS 1?E??ZBΪmU/?~xY'y5g&΋/ɋ>GMGeD3Vq%'#q$8K)fw9:ĵ x}rxwr:\TZaG*y8IjbRc|XŻǿI u3KGnD1NIBs RuK>V.EL+M2#'fi ~V vl{u8zH *:(W☕ ~JTe\O*tHGHY}KNP*ݾ˦TѼ9/#A7qZ$*c?qUnwN%Oi4 =3ڗP 1Pm \\9Mؓ2aD];Yt\[x]}Wr|]g- eW )6-rCSj id DЇAΜIqbJ#x꺃 6k#ASh&ʌt(Q%p%m&]caSl=X\P1Mh9MVdDAaVB[݈fJíP|8 քAV^f Hn- "d>znNJ ة>b&2vKyϼD:,AGm\nziÙ.uχYC6OMf3or$5NHT[XF64T,ќM0E)`#5XY`פ;%1U٥m;R>QD DcpU'&LE/pm%]8firS4d 7y\`JnίI R3U~7+׸#m qBiDi*L69mY&iHE=(K&N!V.KeLDĕ{D vEꦚdeNƟe(MN9ߜR6&3(a/DUz<{ˊYȳV)9Z[4^n5!J?Q3eBoCM m<.vpIYfZY_p[=al-Y}Nc͙ŋ4vfavl'SA8|*u{-ߟ0%M07%<ҍPK! ѐ'theme/theme/_rels/themeManager.xml.relsM 0wooӺ&݈Э5 6?$Q ,.aic21h:qm@RN;d`o7gK(M&$R(.1r'JЊT8V"AȻHu}|$b{P8g/]QAsم(#L[PK-![Content_Types].xmlPK-!֧6 +_rels/.relsPK-!kytheme/theme/themeManager.xmlPK-!Ptheme/theme/theme1.xmlPK-! ѐ' theme/theme/_rels/themeManager.xml.relsPK] /"Qddwwև"Aˠ Q * D!('p+r39=B#E JKLOFSBTUYj\josvwy|h} .Z h: %"$%&*V+bddef JMPQRTUVWXZ\]^`abcdeghjklmoqstvxz{}O d:B@K@W:eQwwzM~^fd!#%(`ladfRgegigggggggX\ KLNOSY[_finpruwy|~___"X8@0(  x4 B S  ?D!D!E!E!E!Eܸ!E\!E!E\!E!E\! E! E\! E! E! E!Eܹ!E\!Eܺ!E\!\\\2]2]=]_```aaaIbIbcccЊЊ#     \\\;]?]?]_```aaaRbRbcccՊՊ#  B*urn:schemas-microsoft-com:office:smarttagscountry-region9*urn:schemas-microsoft-com:office:smarttagsState8*urn:schemas-microsoft-com:office:smarttagsdate8*urn:schemas-microsoft-com:office:smarttagsCity9*urn:schemas-microsoft-com:office:smarttagsplace 8=200948DayMonthYear  > E :B;EzzTZV ] ))+"+$+)+0+3+5+:+=+?+E+I+n,v,,,///////////0445555666666666666>7D7r9y9z999999;;;;;;;;;;;;$<-<2<9<%=)=2=3=8===D=H=Q=T= >>>>>>> >$>(>1>2>4>5>7>8>:><>B>C>E>F>H>M>R>S>U>V>X>Y>????????????????UAYA\AaAdAhAkApAsAuAxA|AAAAA B BBBBB"B'B-B/B2B6BoBoFoJoQoWo\oaonopotoyo}oooooooopppppppppppqqqqq q%q)q/q3q9q=qCqGqPqTq]qaqlqqqxq|qqqqrrrrrrrrrrrrrrrrrrrrrsss sss!s&s.s2s=stttt%t)t0t5t>??@@UAA BHBBBXX7[f[[[[[[\\\#]@]A]m]]]^*_+_Q_``W2'0lR z7`Uf'?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~      !"#%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrsuvwxyz{}~Root Entry F BRData x1Table$WordDocument ӡSummaryInformation(tDocumentSummaryInformation8|MsoDataStorepXBR 4BRLKBBCMVEGMQHHZA==2pXBR 4BRItem  PropertiesUCompObj y   F'Microsoft Office Word 97-2003 Document MSWordDocWord.Document.89q