Chapter. 1 Introduction
Yüklə 1,52 Mb.
|
Summary In this chapter, first part of syntactic annotation, i.e. chunking, has been described. First of all the nation of chunk was dealt which seems to be quite similar to the popular notion of phrase. The distinction between the two has been neatly discussed. Since chunking also is an annotation task, it prerequisites a tagset and annotation tool just like POS annotation. Both the tagset and the tool have been described at length. Each tag has been explained with the help of examples and the entire process of manual chunking of previously built POS annotated KashCorpus has been illustrated with the help of example sentences 42 and 43. The snapshots of the tree viewer have been also given along with the chunked data of the example sentences 42 and 43 in SSF to show how chunk projections are created in the interface and how actually they are stored at the back end in SSF. Further, the linguistic issues raised during the process have been given elaborately with sufficient examples. Finally, the results of the annotation work have been presented. The empirical results have been given in the form of bar-charts which have been also briefly interpreted. The theoretical results have been given in the form of guidelines which covers main decisions that have been taken in order to resolve various issues. Chapter.6 Dependency Parsing of KashCorpus “Unfortunately or luckily, no language is tyrannically consistent . . . All grammars leak.” Adverd Sapir, Language (1921)
As already mentioned in the Chapter three and Chapter four, treebank is a set of machine readable parse trees of natural language, encoding the syntactic, semantic or both types of linguistic information. Dependency treebanks are multi-layered annotation pipelines and at each layer, a separate but related set of linguistic information is annotated, in a manner, so that the tags at the lower level facilitate the annotation at the higher level. The obligatory annotation layers of a dependency treebank include a POS layer, a chunk layer and a relational layer. However, further layers of linguistic information like the morphological or discourse level information can be also added, depending upon the intended utility of the layer in a treebank. For the current dependency treebank, only three layers of linguistic information have been taken into consideration. The first layer contains coarse grained hierarchical POS labels for each token/word of a sentence, as discussed in Chapter four. The second layer contains chunk labels for the clusters of words which have been dealt in Chapter fifth. The third layer contains labels for inter-chunk dependency and non-dependency relations which are dealt in this chapter. In this chapter dependency parsing/annotation of already built chunked KashCorpus has been discussed. Dependency parsing involves labeling head-dependent relations at lower level as well as at higher level. The dependency annotation at lower level has been covered under chunking, hence, intra-chunk, which has been already dealt in previous chapter. The dependency annotation at higher level, i.e. at predicate argument structure level, hence, inter-chunk, is the sole concern of this chapter. Section second is concerned with introducing the notion of (deep) syntactic parsing. Section three is concerned with the description of the grammar formalism used as parsing model. Section four is concerned with the description of GRs. Section five deals with the annotation of dependencies. Section six is concerned with the issues raised during the annotation process. Section seven provides the statistical results of dependency annotation. Section eight discusses the inter-annotator agreement and the results that have been obtained in the concerned experiment. Finally, section nine summarizes the chapter.
Generally, parsing refers to syntactic analysis of an input string and parser is a programme that parses an input string automatically. According to Grune and Jacobs (2008), parsers are already being used extensively in a number of disciplines; in computer science for compiler construction, database interfaces and artificial intelligence, in linguistics for text analysis, corpora analysis, machine translation and stylistic analysis, in document preparation and conversion, in typesetting chemical formulae, in chromosome recognition, etc. Although, the term parsing has been derived from Latin phrase paras orationis meaning parts-of-speech, it is a technical term used for manual or automatic grammatical analysis. When the grammatical analysis involves word level analysis, it is called morphological parsing, when it involves phrase or chunk level analysis, it is shallow syntactic or simply shallow parsing and when it involves a clause or sentence level analysis, it is deep syntactic parsing or simply syntactic parsing. Similarly, if the analysis belongs to discourse level, it can be called as discourse parsing. However, in general terms, parsing is a cognitive or computational process of taking an input string and generating some sort of structure for it, e.g. generation of a parse tree for an input sentence. As far as, the end product of syntactic parsing is concerned, it is clear that parsing stays at the heart of treebanking where the syntactic trees are produced by manual or semi-supervised methods. The notion of syntactic parsing is closely linked to the parsing model which provides grammar formalism for determining the nature of output syntactic trees or graphs. As already mentioned in the chapter one and two, there are two main approaches to syntactic parsing. One is based on the popular syntactic notion known as constituency and other is based on the relatively obscure notion of syntax known as dependency. However, the last decade has shown renewal interest in various varieties of dependency grammar, particularly for parsing text corpus and developing dependency treebanks and parsers. The current work is in line with this resurgent dependency wave. The next section gives a brief account of Indian version of dependency grammar.
The goal of Paninian approach is to construct a theory of human communication, i.e. how natural language is used to convey information to the hearer and how the hearer gets on to the intended meaning? Therefore, grammar is seen as the system of rules that establishes correspondence between what the speaker intendeds to say and corresponding utterance s/he produces and also between what the hearer listens and the meaning s/he extracts from it. Paninian Grammar (500 B.C) has been originally written for Sanskrit two and PCG is actually an attempt to interpret Paninian Grammar in new light and apply it to all modern IA languages. According to Kiparsky & Staal (1969), PCG (Bharati et al., 1993) is a variant of dependency grammar. It has been used as parsing model for all treebanks that are being built in India. It is for the same reason that it has been also used in the current syntactic annotation which is final level of annotation in building dependency treebank of Kashmiri. This model helps to capture the syntacto-semantic relations which are instrumental in constructing a sentence. Sentence is considered as a series of modifier-modified relations with a primary modified, main verb (VM), which is the root of dependency stemma (graph or tree). The elements which modify main verb are its arguments and adjuncts that participate in the action specified by the verb. The relations of these participants with the main verb are called karaka. Since, Kashmiri is highly inflectional language; there are clear cut case markers or postpositions (vibaktis) on the arguments and adjuncts that participate in an action/event. Such morpho-syntactic cues can be very instrumental in identifying the relation of arguments and adjuncts with its root. To some extent there is one-to-one relation between the karakas & the case markers/postpositions. However, many constructions found in the corpus defy this expected correspondence. It has been found that such correspondences between karaka & vebhakti along with TAM features are very helpful in syntactic annotation of Indian Languages which are relatively free word-order in nature (ibid). For illustration consider the following sentence: raath dits library nish bAshiir-an farooq-as neelofer-as khA:trI akh kitaab. Yesterday give-PRF library near Bashir-ERG Farooq-DAT Neelofer-DAT for one book Yesterday Bashir gave a book to Farooq for Neelofer near library.
Figure.1. Paninian Dependency Graph In the above sentence, there is an action represented by the finite verb dits (gave) which is also the root of Paninian stemma, shown in Figure.1. Since the verb is ditransitive in nature, it has three valency slots for arguments. Therefore, there are three arguments represented by three NPs in which Bashir is SUB which has agentive role and has kartaa (k1) relation with the root, Farooq is IO which has the semantic role of recipient or beneficiary and has sampradaanaa (k4) relation with the root and kitaab (book) is DO which has the semantic role of patient and has kartaa (k2) relation with the root. Besides, these participating NPs which fill the valency slots of the verb and play the core roles directed by it, there are additional NPs which are external to the predicate argument or sub-categorization frame of the verb and hence, play secondary non-participatory roles. Some of the NPs (which project from NSTs) provide location for the action diyun (to give). The raath (yesterday) provides temporal location therefore, has kaala adhikarnaa (k7t) relation with the root and the library nish (near library) provides spacial location and therefore, has disha adhikarnaa (k7p) relation with the root. However, the NP Neelofer is neither part of sub-categorization frame nor does it stem from NST and hence, doesn’t provide any information related to direct participation or location of an action or an event but represents an indirect participant which is the purpose of the action. Therefore, Neelofer is a purpose NP which has Taadarthya (rt) relation with the root.The dependency labels that have been devised based on karakas are given in the Figure.2. The description of these karakas is given in the section four of this chapter.
Figure.2. Grammatical Relations Shown in HTB Guidelines Keeping aside limitations and strengths of dependency grammar in general, the criticism that is being leveled upon the PCG is that it lacks tight formalism and doesn’t distinguish between arguments and adjuncts. It is fact that there are hardly any syntactic notions like the transitivity or the argument-adjunct distinction either in the original Paninian grammar or in the current PCG. It is because it is essentially a syntacto-semantic theory that has hardly to do anything with the syntactic notions of sub-categorization, argumentation and adjunction but it must be noted that the syntactic categories like argument and adjunct can be easily extracted from the dependency labels itself. Further, the notion of karaka are roughly equivalent to the notion of semantic role but the karaka relations are identified through the notions semantic roles, subject and object, otherwise, unless one has complete hold on Sanskrit, it is impossible to know what a particular karaka is all about.
So far parts-of-speech tagset and chunk tagsets have been described in chapter four and five, respectively which were one of the pre-requisites to carry out the annotation at the respective levels. Similarly, in order to carry out annotation at this level, i.e. syntactic annotation, an inventory of grammatical relations (GRs) is needed. The following table presents the set of the GRs that have been used in developing the current treebank. These GRs, essentially Sanskritic, are given along with their interpretations and the attachment labels and their variants that have been used at this level of dependency annotation.
Table.1. Showing Grammatical Relations in KashTreeBank The main twenty five GRs, given in the Table.1, though used in developing the current dependency treebank of Kashmiri and other ILs (Hindi, Urdu, Telegu and Bangla) are not all dependency relations. Many of them are no-dependency in nature and are very important for accounting the structure of a sentence. These GRs, dependency or non-dependency, can be divided into eight types, depending up on the nature of relation they represent. The type wise description of the GRs (dependency and non-dependency and karaka or non-karaka) is given below with detailed examples, so that this description would also serve as guidelines for the dependency annotation:
These include all six karaka relations which are the core of Ashtaadhyaayii. They include Karta labeled as k1, pk1, jk1 and mk1, Karma labeled k2, k2g and k2p, Karna labeled as k3, Sampradaana or Anubhava Karta labeled as k4, k4a and k4v, Apadana labeled as k5 and k5prk, and Adhikarana labeled as k7, k7t and k7p. Type one GRs also include three non-karaka relations such as Karta-samanadhikarana and Karma-samanadhikarana labeled as k1s and k2s, respectively and Samanadhikarana labeled as rs-k1 and rs-k2. The description of each Type-one grammatical label (GR) is given below:
It is the most independent of all karakas. The chunk or clause having k1 relation with the finite chunk is generally subject with agentive role but there are non-agentive instances also. Therefore, karta can be either primary karta which is volitional in nature or secondary karta which is non-volitional in nature. In nominative constructions, it is SUB bI chu-s batI khey-vaan. (1) I-NOM be-PRS rice eat-PROG I am eating rice.
I-ERG eat-PRF.SG.MAS rice I ate rice. However, there are exceptional cases which may not fall under the aforementioned criteria. In such cases the SUB may be marked with a case which is dative by form but not by function. For example: feroz-as pazi-nI zyaadI davIdav karin’* (3.a) Feroz-DAT need-NEG more struggle do-INF.SG.FEM Feroz needs not to struggle hard. *farooq-as chi yi kitaab parIn’ (3.b) Farooq-DAT be-PRS-SG-FEM this book read-INF.SG.FEM Farooq has to read this book. Here also, the DREL between the SUB and the verb is marked as k1, irrespective of the fact that it is non-volitional as mentioned above that karta can be volitional or non-volitional. Therefore, it must be noted that volitionality (agentiveness) is not the sole criterion for being karta of a verb but it is fact that it is the strongest criterion. In Kashmiri, passives are formed by a combination of an infinitival oblique verbal form -nI and a periphrastic auxiliary yun (to come) in perfective form, in which the internal argument of the transitive verb surfaces as the subject of the sentence. The agent of the action is not overtly realized and preferably omitted. Therefore, the agentive phrase is optional. However, if the agent is realized, it is either in the form of -zaryi or -athi phrase (a kind of by phrase). farooq-an khuul kuluf. (ACTIVE VOICE) (4) Farooq-ERG open-PRF lock Farooq opened the lock.
Farooq-GEN by come-PRF lock open-PASS The lock was opened by Farooq.
Like any other morphologically rich language, causative or double causative verbs in Kashmiri are formed by morphological process, by suffixing –Inaav in single causatives where there is a causer and a causee, and by doubling the suffix –Inaav in double causatives where, in addition to a causer and a causee arguments, there is one more argument (NP chunk) called mediator causer. The Prayojaka Karta is the causer NP, Prayojaya Karta is the causee NP and Madhyastha Karta is the mediator causer. The dependency relations which causer, causee and mediator causer NPs hold with the causative verb (root/head) are labelled as pk1, jk1 and mk1, respectively, as shown below, in (6), (7) and (8), respectively. For example: arshid-an dyaav-InA:v feroz-as athi aijaaz-as kitaab. (6) Arshid-ERG give-PRF-CAU.SG.FEM feroz-DAT by aijaaz-ACC book Arshid made Feroz to give book to Aijaaz.
Arshid-ERG give-PRF-CAU.SG.FEM feroz-DAT by aijaaz-ACC book Arshid made feroz to give book to aijaaz.
aijaa-zas akh kitaab. (8) Arshid-ERG give-PRF-CAU-CAU.SG.FEM feroz-GEN through shaanuv-DAT by aijaaz-ACC a book Arshid made feroz give aijaaz a book through shanuv. It is evident from the above examples that pk1 is the DREL of ergative marked NP chunk (causer), jk1 is the DREL of –athi marked NP chunk (cause) and mk1 is the DREL of –zAryi marked NP Chunk (mediator causer) with the root of the sentence, i.e. causative verb.
It is the DREL which predicative JJ chunk holds with the verb. These JJ chunks function as SUB compliments. The NP chunks at the predicative position in the copular constructions can be also considered as SUB compliments and can be labelled as k1s. For example: farooq chu baDi neyk. (9) Farooq be-PRS-SG-MAS very pious Farooq is very pious. farooq chu scholar. (10) Farooq be-PRS-SG-MAS scholar Farooq is a scholar.
The OBJ NP chunks which have semantic role of patient are the Karma and the DREL they bear with their heads are labelled as k2, irrespective of whether the construction is in active or passive configuration. The finer distinctions have been also drawn within karma and have been labelled as k2p and k2g which are dealt separately. In ergative constructions (active voice), it is OBJ (k2) which agrees with the verb, as shown in (12). Also in passive configurations, the agreeing NP chunk is karma as shown in (15). In short, the agreeing NP chunks in ergative constructions are karma. These are actually unmarked (accusative) OBJs in both nominative and ergative constructions and the DREls they hold with the root are labelled as k2. However, there are instances where a whole clause introduced by a subordinating particle can be treated as the OBJ. It is called Vakya Karma or clausal OBJ and is also labelled as k2, as shown in (13). For example: farooq chu kitaab par-aan. (11) Farooq be-PRS-SG-MAS book read-PROG Farooq is reading a book. farooq-an chi yim-I kitaab-I pArm-ItsI. (12) Farooq-ERG be-PRS-SG-FEM this-Pl book-Pl read-PRF-PL.FEM Farooq has read this book.
Farooq-ERG say-PRF that he-DAT be-PRS-SG-MAS-NEG no one impress-PROG-EMP Farooq said that no one impresses him. yuh-us A:y’ maah-i-ramzaan-as manz vaariyaa khAzIr bAgraavnI. (15) This year come-PRF-PL-MAS in ramazaan-DAT lot of dates distribute-PASS This year in Ramadan lot of dates have been distributed.
The OBJ NP chunks which have semantic role of goal or destination is also Karma but the DREL they bear with their heads are labelled as k2p, irrespective of whether the construction is in active voice or passive conformation. For example: farooq-as chu garI gatsh-un. (16) Farooq-DAT chu-PRS-SG-MAS garI gatsh-INF Farooq has to go home. farooq gatsh-i garI. (17) Farooq go-FUT home Farooq will go home.
In the sentences with ditransitive verbs where there are no giver-recipient roles, the second OBJ is called secondary karma and bears k2g DREL with the root. For example: yim pagal lukh chi gulzaaras piir sA:b vanaan. (18) These insane people bePRS-Pl gulzaar-DAT saint call-HAB These insane people call Gulzar a saint. It must be noted the semantics of verb vanun (to call) presupposes that there is a person who calls, a thing/person that is to be called and another the name by which the thing has to be called. All the three presuppositions are nominal in nature but can’t be attributive in nature.
The difference between k2g and k2s seems to be very confusing given the fact that in both cases the verbs involved, e.g. vanun (to say) and maanun (to believe), are ditransitive in nature. However, in the latter case, as illustrated in (19) and (20), the predicative JJP or NP (in copula constructions) can’t be treated as arguments but OBJ compliments. The reason for treating them as compliments is that they are attributive in nature, carrying the attributes of OBJ, rather than being nominal in nature like OBJ, so that they could be treated as arguments. These are just like SUB compliments. For example: arshid chu shaanuv-as rut samjaan. (19) Arshid be-PRS.SG.MAS shaanuv-ACC nice think-HAB Arshid thinks that Shanuv is nice. chiin chu hindostaan-as askh taaqathvar muluk maanaan. (20) Chiin be-PRS-MAS hindustaan-ACC a strong country consider-HAB China considers India a strong country.
In the sentences with transitive verb, Karna is that NP chunk through which the action has been carried out by the agent or which is instrumental in carrying out the action. The instrumental role exists irrespective of the type of sentence configuration, i.e. whether the sentence is in active, passive or WH configuration. However, it should be noted that it is not part of argument structure like the causee in the causative constructions. The DREL of instrumental marked NP chunks has been labelled as k3. In short, –sI:t’ marked NPs are Karna but –sI:t’ is ambiguous and shows syncretism between instrumental and associative roles. For example: farooq-an khuul kunz-i sI:t’ kuluf. (21) Farooq-ERG open-PRF with key-ABL lock Farooq opened lock with key. maaji aaprov bachch-as chamchi sI:t’ batI (21) Mother-ERG feed-PRF.SG.MAS kid-ACC spoon-ABL with rice The mother has feed kid rice with spoon.
Lock come-PRF.SG.MAS key-ABL with open-PASS The lock was opened with key.
The OBJ NP chunks which have semantic role of recipient or beneficiary or represent final destination of an action are the Sampradaana. In the sentences with ditransitive verbs, it is the dative marked NP (DO) which is recipient, beneficiary or final destination and holds k4 DREL with the root. It must be noted that the semantic role receiver is not constrained by animacy feature in Kashmiri and even the inanimate OBJs can be marked with dative case, as shown in (26). For example: farooq-an di-ts Suhail-as kitaab parnI khA:trI. (23) Farooq-ERG give-PRF.SG.FEM Suhail-DAT book for reading Farooq gave Suhail a book for reading. tse vAn-ith mea raath rIts kath. (24) you-ERG tell-PRF-2PC me-DAT yesterday nice talk You told me yesterday a nice thing.
I-ERG tell-PRF-1PC.SG.FEM you-DAT yesterday nice talk I told you yesterday a nice thing.
They locked the door. sw kitaab chi dAh-an rop-yan yiv-aan. (27) That book be-PRS.SG.FEM ten-DAT rupee-DAT.PL.FEM come-PROG That book cots ten rupees.
The SUB NP chunk which has the semantic role of a passive experiencer is Anubhava Karta, who perceives through a process represented by the (intransitive) verb. In the clauses with perception verbs, it is the perceiving entity. However, it must be noted that perceiving entities in Kashmiri are not constrained by animacy feature, as shown in (32). In short, it is the dative marked SUB NP chunk which is experiencer and have k4a DREL with the root. For example: mea lAj bochi. (28) I-DAT.SG hurt-PRF.SG.FEM hunger I felt hungry. lADk-as peyi nindIr. (28) boy-DAT.SG.MAS fall-PRF sleep The boy felt asleep. lADk-an tor fiqri (29) boy-DAT.PL.MAS cross-PRF.SG.MAS understand The boys understood. kor-i gov shakh. (30) girl-DAT.SG.FEM go-PRF.SG.MAS doubt The girl got doubt. kor-eyn aav mushuk. (31) girl-DAT.PL.FEM come-PRF.SG.MAS smell The girls caught smell. makaan-as peyi traTh (32) house-DAT.SG.MAS fall-PRF.SG.FEM lightening The house is struck by thunder. dukaan-as log naar. (33) shop-DAT catch-PRF.SG.MAS fire The shop caught fire. tsuunT-is log daag. (34) apple-DAT catch-PRF.SG.MAS stain An apple caught stain. makaan-as peyov pash vAs’ (35) house-DAT fall-PRF.SG.MAS roof The roof of the house fell down.
It is a unique kind of Anubhava Karta specific to Kashmiri which has a different semantic role to play which is entirely different from the recipient/ beneficiary/destination or experiencer. This role is sort of possessor-experiencer with a clause structure like that of k1s, i.e. the type of clause structure of predicative adjectives and that of copulas which simply state the existential state of affairs. The DREL which is labelled as k4v actually holds between a dative SUB, showing some existential state of affairs, and the root. For example: farooq-as chi sath neychiv (36) Farooq-DAT be-PRS.PL seven sons Farooq has seven sons. bAshiir-as chu sharaarath. (37) Bashir-DAT be-PRS.SG.MAS anger Bashir is angry. farooq-as chu-nI heys-Iy. (38) Farooq-DAT be-PRS.SG.MAS-NEG consciousness-EMP Farooq is unconscious. bAshiir-as chu safeyd mas. (39) Bashir-DAT be-PRS.SG.MAS white Bashir has white hairs.
The NP chunk which is the source of an activity, i.e. the point of departure or starting point of an action or activity, is the Apaadaana. In Kashmiri it is the ablative marked NP followed by ablative marked location postposition, which represents source or starting/departure point of an activity. It should be noted that the cases like (41) and (42) might appear confusing. However, the former ablative marked one is right example of k5 but not the latter dative marked one. The difference lies in the ablative marked and the unmarked locative postpositions, i.e. peyTh-I and peyTh, which is enough to recognize that the former case is an instance of Apaadaana while the latter can be the instance of simple locative. Although, semantically it is obvious that in former case, there is a sense of departure (ablative effect) which is lacking in the latter case but it would be semantically anomalous to consider latter case a simple spatial locative adverbial as the ‘plate’ in (42) can be instrumental or even ablative but can’t be space where the action of drinking takes place. The DREL which such ablative marked NPs hold with the root of the clause is labelled as k5. For example: kul-i peyTh-I pyov Duun pathar.* (40) Tree-ABL on-ABL fall-PRF.SG.MAS walnut down A walnut fell sown from the tree. farooq-an cheyi kawl-i manz-I treysh. (41) Farooq-ERG drink-PRF.SG.FEM bowl-ABL in-ABL water Farooq drank water from the bowl.
Farooq be-PRS.SG.MAS room-DAT.SG.MAS in plate-DAT.SG.MAS in rice eat-PROG Farooq eats rice in plate in the room. farooq-as nish-I draav su raath. (43) Farooq-DAT near-ABL left he yesterday He left from Farooq (Farooq’s place) yesterday. su nafar tsol az militaryvaaly-an nish-I (44) That man run-PRF.SG.MAS today military man-DAT.SG.MAS
That man ran away today from military. There are cases where the activity represents a process through which change of state of a substance occurs. In such cases, the source point from which that change starts is the NP representing one substance (raw/natural material) which changes into another substance (the product), represented by another NP. The former NP which undergoes change is said to be the Prakruti Apaadaana and its DREL with the root is represented by a variant of k5, i.e. k5prk. The dative marked NP which represents base substance or raw material is the Prakruti Apaadaana. For example: guur chu dod-as tsaaman banaavaan. (45) Milkman be-PRS.SG.MAS milk-DAT.SG.MAS cheese make-PROG Milkman makes cheese out of milk. sIts chu kapr-as palav suvvaan. (46) Tailor be-PRS.SG.MAS cloth-DAT.SG.MAS dress-Pl sew-PROG Tailor makes dress out of cloth.
The NP chunk which denotes a location in space for an action, involving different participants like Karta, Karma or Karna, is called Desha Adhikarana. These are not only those cases which are typical spatial locatives and are tagged as NST at POS level, e.g. yetyth (here), hotyth (there), etc but also those cases which are typical nouns and are tagged as NN/NNP at POS level, e.g. garI (home), saDakh (road), kuTh (room), etc. However, both are NPs at chunk level, therefore, any NP which provides spatial location for an event, action or a state is Desha Adhikarana. It is generally a dative marked NP followed by postposition indicating spacial location. The combination of –DAT marker and locative postposition, e.g. - as peyTh, in Kashmiri has corresponding complex postposition in Urdu/Hindi, e.g. ke uupar. It has been found that whatever role genitive (ke) performs in the formation of Hindi/Urdu complex postpositions (where it ceases to be genitive and the postposition projects itself in non-compositional way), the dative markers (-i/-yan/-as/-av) in Kashmiri perform the same role but it can’t form a complex postposition as the dative itself is merely a marker (a bound form) unlike genitive of Urdu/Hindi which occurs as postposition (a free form). However, the –DAT in Kashmiri also ceases to be dative when it is followed by locative position and the GREL it represents has totally locative interpretation as if the markers, –i, –yan, –as, and –av are no more dative markers but mere obliqueness markers. It must be noted that the dative markers are very controversial in Kashmiri Koul and Wali (2006) treats them as –DAT but Emily Manetta (2008*) treats them as obliqueness markers. The DREL of Desha Adhikarana with the root of clause is labelled as k7p. For example: tati Os Farooq kursy-i peyTh bih-ith. (47) there be-PST.SG.MAS Farooq be-PST.SG.MAS chair-DAT.SG.FEM on sit-PART There Farooq was sitting on the chair. su chu bon-i tal aaraam karaan. (48) He be-PRS.SG.MAS chinaar tree-DAT.SG.FEM under relax do-PROG He is relaxing under the Chinar tree. farooq chu kuTh-is manz paleT-as manz batI kheyvaan.* (49) Farooq be-PRS.SG.MAS room-DAT.SG.MAS in plate-DAT.SG.MAS in rice eat-PROG Farooq eats rice in plate in the room.
pleasant remain-PROG In Mysore weather remains very pleasant.
The NP chunk which denotes location in time for an event, action or an activity involving various participants is called Kaal Adhikarana. These not only include the cases which are typical temporal locatives and are tagged as NST at POS level, e.g. vun’ (now), patI (latter), etc but also those cases which are typical nouns and are tagged as NN/NNP at POS level, e.g. shaam-as (in the evening), pagah (tomorrow), 1950-has manz (in 1990), etc. Both the cases will be NPs at the chunk level which have adverbial function. Such chunks which provide the temporal location of an event or an action are usually dative marked NPs. For example: subh-as gayizi garI panun. (51) Morning-DAT go-2PC home own In the morning go your own home.
1947-DAT in go-PRF.SG.MAS india free In 1947 India got freedom.
Yesterday reach-PRF.SG.MAS Farooq home five O’clock Farooq reached home yesterday at five O’clock.
then rise-PRF.SG.MAS he-DAT fever Then he caught fever.
The NP chunk which denotes location of an event, action or an activity elsewhere, i.e. other than concrete space and time is called Vishaya Adhikarana. Generally, the NPs which are non-spatial and non-temporal in nature are marked with dative and followed by locative postposition like Desha and Kaal Adhikarana. Such NPs include abstract entities that capture some notion of abstract space. The DREL such NP holds with root is labelled as k7. For example: kejriiwaal chu az kal surkhi-yan manz aasaan. (55) Kejriwal be-PRS.SG.MAS today tomorrow headline-DAT.Pl.FEM in remain-PROG These days Kejriwal continuously remains in the headlines.
My talk-DAT.SG.FEM on climb-PRF he-DAT.SG.0 very anger He became very angry on my argument. siyaast-as manz vasun chu-nI Thiekh. (57) Politics-DAT in enter-INF be-PRS.SG.MAS -NEG good To enter into politics is not good.
He-GEN-DAT.SG.MAS idea-DAT.SG.MAS in be-PRS.Pl.MAS own-GEN.SG.FEM good In his opinion everyone is correct at his place.
It only includes only two GRs, r6 and rsp. These are non-karaka but dependency GRs which holds between two nominals, i.e. between two nouns or between a pronoun and a noun. It rarely occurs between two pronouns. The description of r6 and rsp DREls are given below.
The NP chunks which have genitive relation with the other NP chunks are the Shashthi. In Shashthi GR, there are like any other dependency relation a dependent and a head, the possessor and the possessed. The first NP is usually possessor and the second NP is possessed and it is the possessed NP which is head and possessor the dependent. This is the only GR that holds between two nominals and not between a nominal and a verb root. For example: farooq-un boD bOy chu polices-as manz. (60) Farooq-GEN elder brother be-PRS.SG.MAS in the police-DAT Farooq’s elder brother is working in police. tAm’ sInz majbuurii ma kariv nazar andaaz. (62) He of-SG.FEM predicament not do-Pl ignore Do not ignore his predicament. kaam-i hInz jaldii kin’ draav su garI. (63) work-ABL.FEM of-SG.FEM hastiness Because of urgency of work he went home. yeti-ch sadakh chi vaariyah kharaab. (64) here-GEN road be-PRS.SG.MAS very bad Road of this place is very damaged. makaan-uk pash peyov vAs’. (65) house-GEN roof fall-PRF down The roof of the house fell down. The label For example:
Human hand can do many types of work. In the above sentence (Q.6), the first genitive marked NP is dependent on the NP which having k1 relation with root of the sentence and the second genitive marked NP is dependent on the NP which has k2 relation with the root.
The NP chunks which indicate the duration (temporal) or span (spatial) of an event, an action or a state is durative expression. There are two points in durative expressions, viz. a point of starting and an end point. The duratives, consisting of two NPs, may function as temporal, spatial or manner adverbs. The starting point NP depends upon the end point NP and the DREL between them is labelled as 1982(-I) peyTh-I 2012-as taam ruudus bI baDI khosh. (67) 1982(-ABL) on-ABL 2012-DAT to remain-PRF-1PC.SG.MAS I very happy From 1982 to 2012 I remained very happy.
Kashmir-ABL on-ABL Kanyakumari to be-PRS.SG.Fem-NEG any-Emp train There is no train from Kashmir to Kanyakumari.
These include rd, rh, rt, ku, ras, rsp, and rad GRs which are also non-karaka but dependency relations holding between the dependent NP and the root of the clause like Type-one GRs but unlike the Type-two GRs which holds between a dependent NP and non-root head. The description of these relations is given below:
The NP chunk which indicates the direction of an activity is the Prati or the directional NP. In Kashmiri, ‘kun’ is the directional postposition and any NP consisting of directional postposition is Prati. It has been observed that like the NPs consisting of locative postpositions, the NPs which consist of directional postpositions are marked with dative. Also, like locative NPs, the directional NPs are mere adjuncts. It must be noted that there are certain directionals which do not really show the direction of an activity but show some metaphorical direction as shown in (72). The GR which such NPs bear with the root are labelled as ‘rd’. For example: farooq draav darvaaz-as kun. (69) Farooq go-PRF door-DAT towards Farooq went towards the door. su chu mea kun vuchaan. (70) He be-PRS.SG.MAS me towards see-PROG He is looking/ towards me.
He be-PRS.SG.MAS right towards He moves towards right.
Those people be-PST.Pl.MAS opposite protest do-PROG Those people were protesting against police.
The NP chunk which indicates the reason or cause of an activity is the Hetu or the reason NP. In Kashmiri, (sI:t’ and kiny) are the reason postpositions and any NP consisting of reason postposition is Hetu. However, these NPs need to be clearly distinguished from the instrument NPs which also use (sI:t’) postposition. (sI:t’) is actually an instance of case syncretism in Kashmiri. The GR which such NPs bear with the root are labelled as ‘rh’. For example: tami vajah kiny heyok-nI police-an su band thA:vith. (73) That reason because of can-PRF.SG.MAS -NEG police-ERG he close keep-PRF Because of that reason police couldn’t kept him locked. tami sabb-I gov mea hana tseyr. (74) That reason-ABL go-PRF.SG.MAS I little late Because of that reason I am late.
This cold due to fall-PRF.1P.SG.MAS I too ill I too have fallen ill due to/because of this cold.
I be.PST.SG.MAS pain due to shiver-PROG I was shivering due to pain.
The NP chunk which indicates the purpose of an activity is the Taadarthya or the purpose NP. In Kashmiri, khA:trI, mokhI, muujuub, and baapath (for) are the purpose postpositions and any NP consisting of purpose postposition is Taadarthya. However, sometimes VGNN also performs purposive role as shown in (80) in which a gerund is marked with purposive case. The GR which such NPs bear with the verb root are labelled as ‘rh’. For example: faqooq-ni khA:trI pyov yi soorui karun. (77) Farooq-GEN for fall-PRF.SG.MAS this all do-INF All this needed to be done for Farooq.
Work-ABL for come-PRF.SG.MAS I here I came here for work.
You-GEN for go-FUT she Srinagar For you she will go to Srinagar.
Rice eat-PART go-PRF he sleep-GER-PUR Having eaten rice he went for sleeping.
The NP chunk which indicates the similarity or comparison (expressed through predication) between two entities is the Saadrishya or the comparand NP. In Kashmiri, khotI and nish are the comparand postpositions while as pA:Th’ and hiuv or hish are the Similative postpositions. Hence, any NP consisting of comparand or Similative postposition is Saadrishya. However, the NPs with pA:Th’ postposition need not to be confused with adverbials like Thiekh-pA:Th and also nish postposition need not to be confused with locative postposition. The forms, nish and pA:Th’, are ambiguous and can perform two functions in to contexts. These are actually two more instances of case syncretism in Kashmiri. The GR which such NPs bear with the root can be labelled as ‘k*u’ but if the comparison or similarity is with SUB NP, the GR of comparand NP is labelled as k1u and similarly if it is with OBJ NP, the GR is labelled as k2u. Actually, the star mark (*) can be seen as a variable mark which is substituted with any karaka, depending upon the comparee. For example: koshur treebank chu vuni hindi tI urdu treebank-av khotI/nish vaariyaa lokut. (81) Kashmiri treebank be-PRS.SG.MAS yet Hindi and Urdu treebank-ABL.Pl.MAS as compared to very small Kashmiri treebank is yet very small as compared to Hindi and Urdu treebanks.
Farooq be-PRS.SG.MAS basher-GEN like good person Farooq is a good person like Bashir.
Bashir be-PST.SG.MAS gaint like-SG.MAS person Bashir was a giant like man.
Farooq-DAT feel-PRF.SG.FEM that-SG.FEM girl Nowsheen-DAT like-SG.Fem That girl looked like Nowsheen to Farooq.
The NP chunk which indicates the association of an entity with other entity in performing an activity is the Upapaada-sahakaaraktawa or the associative NP. In Kashmiri, sI:t’, saan, heyth, bagA:r and varA:y are the associative postpositions and any NP consisting of associative postposition is Taadarthya. The sI:t’ is positive associative postposition while as bagA:r and varA:y are negative associative postpositions. The GR which such NPs bear with the verb root can be labelled as For example: su draav pann-is mA:l-is sI:t’ chakr-as. (85) He go-PRF.SG.MAS own-DAT father-DAT with walk-DAT He went on a walk with his father.
India be-PRS.SG.MAS China-DAT talk do-GER-GEN for ready India is ready to talk with China.
Farooq-ERG eat-PRF.SG.MAS apple peel with/along with Farooq ate apple with/along with peel.
Military men be-PST.Pl.MAS weapons-ABL without roaming Soliders were roaming around without weapons.
The NP chunk, which indicates addressing to some person, bears a DREL with the verb root which is labelled as moj-ai, mea di-tay batI. (89) Mother-VOC.SG.MAS me give-2PC.SG.FEM rice. Mother, give me rice.
Farooq-VOC.SG.MAS tala-MOOD here come Farooq, you come here.
Sir, he say-PRF that I go-FUT.1PC.SG home Sir, he said that he will go home.
Hey-SG.MAS that book finish-WH.SG.MAS read-PART Hey! You finished that book.
The NP chunk, which indicates the source of an information or point of view which may or may not be a person, is information source NP. In Kashmiri, the NP consisting of genitive marked nominal and informational postpositions (mtA:bik and hisaabI) bear a DREL with the verb root which is labelled as farooq-ni mutA:bik pazi mea garI vaapas yun. (93) Farooq-GEN according to should I home back come-GER According to Farooq I should come back home.
You-GEN way-ABL be-PST.SG.MAS he lie tell-PROG According to you he was telling a lie.
The NP chunk, which indicates an entity towards which all the information of a proposition is directed, is information target NP. It can be also a clause. In Kashmiri, the NP consisting of dative marked nominal and the target postposition mutliq (about) bears a DREL with the verb root which is labelled as farooq-an vAn’ mea bAshiir-as mutliq akh kath. (95) Farooq-ERG tell-PRF.SG.FEM me Bashir-DAT about one talk Farooq told me something about Bashir.
The NP chunk, which indicates a hurdle in an activity that has been overcome, is the hurdle NP. The hurdle can also be a whole clause. In Kashmiri, the NP consisting of genitive marked nominal and hurdle postposition baavojuud (in spite of or despite) bears a DREL with the verb root which is labelled as tAm’sIndi inkaar karnI baavojuud pyo mea tor gatshun. (96) He-GEN denial in spite of fall-PRF.SG.MAS I go-INF In spite of his denial, I had to go there.
Difficulty-Pl.FEM in spite of keep-PRF.SG.MAS -1PC he own work do-PROG In spite of difficulties, he kept doing his work.
Type-Four relations are essentially to capture the adverbial relations of inherent adverbs and participles. Since, the same participles also modify nouns; participial noun modifiers are also included in this class of GRs. It includes adverbials (adv and sent-adv), vmod (vmod_Rh and vmod_Inst) and nmod.
The manner adverbs which are discontinuous and hence, didn’t form a chunk with verb but gave rise to their own chunk projections tagged as RBPs depend on the verbal root and the DREL they bear with the root is labelled as su draav vaarI vaarI garI kun. (98) He leave-PRF.SG.MAS slowly towards home. He left towards home slowly. tim chi yim-I kath-I baar baar karaan. (99) They be-PRS.Pl.MAS these talk-Pl again do-PROG They talk it again and again.
This like should-2PC-NEG these talk-Pl do-INF.Pl You shouldn’t talk like this.
The participial forms which may or may not constitute non-finite clauses are projected as chunks and are attached to the root in order to show that they bear modifier relation with it. In Kashmiri, the –ith and –aan forms of verb are participial forms and constitute VGNF chunks. The –ith forms like shongith (having slept), bihith (having sit), tsA:p’ith (having chewed), etc may be sequential, consequential or instrumental in nature as far as their role is concerned while as the –aan forms like pakaan pakaan (during walking), gindaan gindaan (while playing), etc which are actually progressive/habitual forms but express simultaneity on reduplication and function as verb modifiers. The –ith forms are also reduplicated but when reduplicated they change their form like shongith (having slept) changes into shong’ shong’ to encode simultaneity and sequentiality together in more complex way in order to express manner. The DRELs which all these variants of VGNF bear with the root are labelled with an underspecified label batI khe-yth draav su shong-ni. (101) Rice eat-PART go-PRF he sleep-GER-PUR Having eaten rice he went for sleeping. mAr’-mAr’ chu su kA:m karaan. (102) die-PART-RED be-PRS-SG.MAS he work do-HAB He works slowly.
Walk-HAB-RED be-PST.SG.MAS he ice cream eat-PROG While walking he was eating ice cream.
Some participials forms are projected as VGNF chunks like the verb modifier participials as mentioned above but are attached to the modified noun instead of the root to show their modifier relation. In Kashmiri, the –vun, –vol and –ith forms are noun modifier participials. The –vun forms like asvun (laughing) and vudvun (flying), –vol forms like asanvol (laughing) and natsanvol (dancing) and –ith forms like bihith (sitting) and shongith (sleeping) are the noun modifier participials and the DREL they bear with the nouns is labelled as For example: asvun insaan chu saarinIy khosh kar-aan. (104) smile-PART person be-PRS.SG.MAS all happy do-HAB All people like a smiling person.
Play-PART person be-PRS.SG.MAS healthy remain-HAB The person who plays remains healthy.
tree-DAT on sit-PART crow Taav Taav do-PROG The crow sitting on the tree is crowing.
It includes clausal modifications brought about by relative clauses. Relative clauses are embedded clauses introduced by relative pronouns. The relative pronouns have their corresponding pronouns in the matrix clause and therefore, these clauses are called relative–correlative constructions. In Kashmiri, yus–su, yuth–tyut, and yithIpA:Th’–tithIpA:Th’ are relative–correlative elements. Relative clauses modify noun/pronoun, adjective and adverb. The description of such GRs is given below:
In Kashmiri, the relative element which introduces a relative clause to modify a nominal is yus and its correlative is su or any other noun. These relative elements are either the relative pronouns or the relative demonstratives. The DREL which these relative clauses bear with the non-root nominal head are that of noun description which is different from normal nominal modification brought about by inherent adjectives or participials. Nevertheless the name used for noun description is nominal modification but it is labeled differently as su nafar yus otrI gov garI yiyi pagah vaapas. (107) That man who day before yesterday go-PRF.SG.MAS home be-FUT.SG tomorrow return The man who went home day before yesterday will return tomorrow.
That man be-FUT.SG tomorrow who day before yesterday gov. The man will return tomorrow who went home day before yesterday.
I be-PRS.SG.MAS .1PC that-SG.FEM book read-PROG which-DAT on hue and cry be-PST.SG.MAS stand-PRF.SG.MAS I read that book on which there was hue and cry. tAm’ dits tas lADkI-as kitaab yAm’ tas mAnj’. (110) He give-PRF.SG.FEM that-DAT boy-DAT book who-ERG ask-PRF.SG.MAS him He gave that boy book the book who asked him for it. farooq-an lyokh tami qalmI sI:t’ zindagii hund falsafI yami sI:t’ tAm’ mea kA:trI akh chiTh leych-mIts A:s. (111) Farooq-ERG write-PRF that pen with life of philosophy which with he-ERG I for one latter write-PRF.SG.FEM be-PST.SG.Fem Farooq wrote philosophy of life with that pen with which he had written a letter to me.
The relative elements which introduce relative clauses in order to modify deictic adjectives are yuth, yith’, yitsh and yitshI and their corresponding correlatives are tyuth, tith’, titsh and titsI, respectively. For example: su chu huu-ba-huu tiyuth-ui yuth tas panun mwl os. (112) He be-PRS.SG.MAS exactly like-that like-which he-DAT his father be-PST.SG.MAS He is exactly like that like which his father used to be.
like-that-EMP jacket buy-IMP.2PC like-which I buy-PRF yesterday You buy that kind of jacket like which I bought yesterday.
The relative elements which introduce relative clauses in order to modify an adverb or deictic adverb is yIthI-pA:Th’, yithI-kIn’ and yami tAriiqI and their corresponding correlative are tithI-pA:Th’, tithI-kIn’ and tami tAriiqI. For example: tithay-pA:Th’ kAri-zi az ti rut dance yithIpA:Th’ raath kor-uth. (114) Like-that do-IMP.2PC today also nice dance like-which yesterday do-PRF.SG.2PC You dance as nicely today also as you did yesterday.
Like-which she I-DAT tell-PRF like-that had I-DAT do-INF As she told me I had to do like that.
It includes a set of non-dependency relations that hold between two conjoined elements or clauses of equal status and that of between two elements or clauses of unequal status. The former is the coordination relation and the latter is the subordination relations which are basically structural relations to keep the organization of compound, complex, compound-complex and complex-compound clauses intact and have nothing to do with modification of the finite verb but have everything to do with the tying together of one or more chunks on intra-clausal level or one or more VGF chunks on inter-clausal level. The description of these relations is given below.
The CCP chunk which conjoins two chunks within a clause is the head of the conjoining chunks and is attached to the root of the clause in which they occur as they both are in symmetrical relation with each other and thus, both are dependents of the root. However, the CCP chunk which occurs at inter-clausal level is considered as the root of the compound and compound-complex sentences that involve many finite clauses. In Kashmiri, CCPs include tI (and), kinI/yaa (or), etc. It must be noted that sometimes even commas can also functions as CCPs. The GR which two chunks or clauses bear with their respective heads, which in both cases is CCP, is labelled as farooq tI bAshiirI chi dohdish paanIvan’ chob chob karaan. (116) Farooq and Bashir be-PRS.Pl everyday each other hit hit do-HAB Everyday Farooq and Bashir fight with each other. farooq chu lamaan dwchun tI bAshiirI chu lamaan khovur. (117) Farooq be-PRS.SG.MAS pull-HAB towards right and Bashir be-PRS.SG.MAS pull-HAB towards left. Farooq pulls towards right and Bashir pulls towards left.
Instead of symmetrically conjoining chunks and clauses, some CCP chunks introduce new chunks new clauses and thus, enter into embedding phenomenon by joining two chunks or clauses asymmetrically. The embedded clauses introduced by complementizers are considered as OBJs already discussed but the rest of the rooted complementizer clause attaches to the complementizer which acts as the syntactic head of the compliment clause but not the VGF. Therefore, the complementizer, e.g. zi and ki (that), is the dependent (OBJ) of the root of matrix clause but simultaneously, it is also the head of complementizer clauses as shown in Fig. which is the graphic representation of (118). The GR the VGF of a complementizer clause bears with the complementizers is also labelled as ccof as shown in Fig.5. For example: farooq-an chu vanaan zi tas chu-nI kahn-ti veytsaan. (118) Farooq-ERG be-PRS.SG.MAS say-HAB that he-DAT be-PRS.SG.MAS -NEG noyone-EMP impress-HAB Farooq says that no one impresses him.
Figure.3 Showing Intra-clausal ccof in (116) 
Figure.4 Showing Inter-clausal ccof in (117) 
Figure.5. Showing Sub-ordinating ccof in (118)
It includes entirely different a set of GRs, pof and fragof, which are actually innovations to handle certain crucial phenomenon like complex predication and v2 phenomenon. Without these relations, it would have been difficult to account for the structures involving such phenomenon. The description of these relations is given below: Yüklə 1,52 Mb. Dostları ilə paylaş:
|
