Language/OS - Multiplatform Resource Library

home *** CD-ROM | disk | FTP | other *** search

/ Language/OS - Multiplatform Resource Library / LANGUAGE OS.iso / icon / contrib / pcw.lha / chap2.grk < prev next >

Wrap

Text File | 1991-11-18 | 71.9 KB | 1,851 lines

.KF:chap2.toc .KW:59 .N:14 .XT:2 .XB:0 .X:10 .L:59 .M:1 L---+----1----+----2----+----3----+----4----+----5T---+---R6----+----7----+----8 .H: .H: .H: .F: .F:...$$$... .M:1 CHAPTER 2 CLAUSE LEVEL ANALYSES .K:2. CLAUSE LEVEL ANALYSES .M:2 This chapter will discuss the grammatical analysis of some selected clauses taken from the Greek text of Matthew 26. Five different analyses will be done with a different theoretical approach used for each. The focus of each analysis will be the relationships of words to each other within clauses. For each of the theoretical ap proaches there will be a discussion of the suitability of that approach for use in machine translation. .H: .H: $$$ .H: .F: .F: All the text to be analyzed is taken from the third edition of the United Bible Societies Greek New Testament (Aland 1975) which has Greek text equivalent to the twenty- sixth edition of the Nestle-Aland Novum Testamentum Graece (Nestle 1979). This is the Greek text of the New Testament which is most widely accepted today. The English translations, the morphological parsings, and the syntac tical analyses are my own. It is not the purpose of this chapter to produce an exhaustive analysis of New Testament Greek syntax utilizing any of the five grammatical theories. That would be a 15 formidable undertaking indeed. Rather, an attempt will be made to give the flavor of what such an analysis might look like in each of the grammatical theories. This will provide a basis from which to make a subjective comparison of the utility of each of the different analyses versus an analysis from one of the other theoretical frameworks. The utility of each analysis will be judged on the basis of its usefulness in developing a machine translation program. In each section the text and its analysis according to one of the grammatical theories follows a few intro ductory remarks. Each text is given following the number of the verse from which it was taken. Under each Greek word its primary sense in English has been given. Beneath the sense in English two lines are dedicated to the morpho logical parsing of each word. These parsings are my lin guistic analysis of each word in isolation. The types of information represented by these tags are case, number, gender, and so on. The abbreviations used in these morphological tags are found in Appendix A. Following the morphological tags and set off by blank lines before and after are the grammatical tags. These describe the syntactic function of each word as it is used in the clause or sentence in which it occurs. The abbreviations used in these grammatical tags are also found 16 in Appendix A. The last entry for each line of Greek text is its free translation in English. Generative Transformational Grammar .K: Generative Transformational Grammar In this section an analysis will be done using the Generative Transformational theory of grammar proposed by Noam Chomsky in his Aspects of the Theory of Syntax (1965). Portions of five verses from the Gospel of Matthew chapter 26 were chosen for analysis. Each of these texts was selected because it contained a Greek verb which could be translated into an English modal. Thus, while the syntactic analysis of the Greek text will not be overly complex, it will be more interesting than the analysis of simple verb phrases. It should be pointed out that not all English modals have an equivalent in New Testament (Koine) Greek. Where English needs the modal will to express action or state of being in the future, Koine Greek has an explicit future tense suffix. Likewise, where a speaker of English would use the modals might or may to express a possibility, Greek has the subjunctive and optative moods to serve this purpose. Nevertheless, Koine Greek does have verbs which 17 express ability (can/could) or obligation (must/should), and these are used in much the same way as English modals. The analysis is limited to the data which follows. No doubt the reader who is familiar with New Testament Greek syntax will notice aspects of even this brief data that could stand further elucidation. An obvious candidate for further analysis is the placement of the conjunction gar 'for' in what is known as postpositive position. However, treatment of this very interesting aspect of Greek syntax would require analysis of more data, and is outside the scope of this section. Greek Data With Analysis .K: Greek Data With Analysis .M:1 9) edunato g a r touto praqhnai it could have been for this to be sold V-Impf-M/P Sub-Conj Dem-Pro V-Aor-Pass Ind-3S Nom-Neu-S Inf Mod Conj NP Inf For this could have been sold. 40) ouk isxusate mian wran grhgorhsai not you were able one hour to watch Adv V-Aor-Act Adj-Acc N-Acc V-Aor-Act Ind-2P Fem-S Fem-S Inf Neg Mod [ NP ] Inf 18 met emou with me Prep Pro Gen Gen-1S [ PP ] Were you not able to watch one hour with me? 42) ou dunatai touto parelqein not it can this to pass away Adv V-Pres-M/P Dem-Pro V-Aor-Act Ind-3S Nom-Neu-S Inf Neg Mod NP Inf This can not pass away. 54) outwj dei genesqai thus it must to be Adv V-Pres-Act V-Aor-Mid Ind-3S Inf Man Mod Inf Thus it must be. 61) dunamai katalusai t o n na o n I can to destroy the temple V-Pres-M/P V-Aor-Act Def-Art N-Acc Ind-1S Inf Acc-Masc-S Masc-S Mod Inf [ NP ] tou qeou the of God Def-Art N-Gen Gen-Masc-S Masc-S [ NP ] I can destroy the temple of God. 19 Summary Of Grammatical Tags .K: Summary Of Grammatical Tags 9) Mod Conj NP Inf 40) Neg Mod NP Inf PP 42) Neg Mod NP Inf 54) Man Mod Inf 61) Mod Inf NP NP Phrase Structure Rules .K: Phrase Structure Rules S -> (Man) Cl Cl -> (Neg) Mod (Conj) (NP) Inf (NP) (NP) (PP) NP -> {Dem-Pro } {Adj Noun } {Def-Art Noun} PP -> {Prep Pro} Man -> outwj Neg -> {ou } {ouk} Mod -> {dei 'it must' } {dunamai 'I can' } {dunatai 'it could have been'} {edunato 'it could be' } {isxusate 'you were able' } Inf -> {genesqai 'to be' } {grhgorhsai 'to watch' } {katalusai 'to destroy' } {parelqein 'to pass away'} {praqhnai 'to be sold' } Dem-Pro -> touto 'this' Adj -> mian 'one' Noun -> {wran 'hour' } {na o n 'temple'} {qeou 'of God'} Def-Art -> {ton 'the'} {tou 'the'} 20 Prep -> met 'with' Pro -> emou 'me' Conj -> gar 'for' Sample Phrase Markers .K: Sample Phrase Markers 40) S | | Cl | | ----------------------------------------------------- | | | | | | | | | | Neg Mod NP Inf PP | | | | | | | | | | ouk isxusate ------------ grhgorhsai --------- not' 'you were able' | | 'to watch' | | | | | | | | | | Adj Noun Prep Pro | | | | | | | | mian wran met emou 'one' 'hour' 'with' 'me' 21 54) S | | --------------------- | | | | Man Cl | | | | outwj --------------------- 'thus' | | | | Mod Inf | | | | dei genesqai 'it must' 'to be' .M:2 Critique .K: Critique This subsection will attempt to answer the question, "To what extent would a Generative Transformational Grammar (GTG) analysis be useful in implementing a machine trans lation program?" The first deficiency in GTG as a basis for implementing machine translation is that it lacks transformations from one language to another. All of the transformations specified by the phrase structure rules of a GTG analysis are transformations between phrase markers at various levels of structural abstraction. However, none of the transformations crosses the boundary between one language and another. Some augmentation of the theory will need to be attempted if it is to be used successfully to translate sentences of one language into that of another. 22 A second weakness in Generative Transformational Grammar is that it says little about semantics. For instance, in verse 40 what is the referent of mian wran 'one hour'? Is it a span of time or some object tangible or intangible to be watched? The answer may be obvious to the linguistic analyst, but not as a result of the GTG theory. Finally, Generative Transformational Grammar is monodirectional, being synthetic rather than analytic. That is to say, GTG specifies how to construct well formed sentences in a language. It does not specify how to parse well formed sentences into their constituent parts. GTG was never intended to be a model of production and recep tion but of a speaker's knowledge of his language. How then might a GTG analysis of a span of text be used in the machine translation of that text into another language? First of all, the parsing of the span of text into its constituent parts would have to be done by hand-- by a human analyst. This procedure would be burdensome only for translation into a single language. In such a case, it would be simpler and quicker for the human analyst merely to translate the text by hand. However, in the case of texts which one desires to translate into multiple languages this objection shrinks in importance. 23 Secondly, any semantic analysis of the text would also have to be done by a human analyst. Any ambiguities would have to be resolved prior to presentation of the data to the machine translation program. For instance, words with multiple meanings would need to be indexed in some way to eliminate the ambiguity. Next, it would be necessary to construct a lexicon which maps words of the source language into words of the target language. The GTG analyses of the two languages would be of some help here since lexical items of a language are represented in features in a Generative Transformational Grammar. Actually, as mentioned above, the lexical items of each lexicon would need to be indexed in order to avoid ambiguities when words have more than one meaning. The end result of mapping these indexed words between the two lexicons would be that the mappings would be between meanings rather than words. There are addi tional problems to be addressed in the semantic and lexical areas such as implicit information, lexical skewing, and the use of metaphors, idioms, and other figures of speech. These problems will be taken up in the chapter on semantic structure analysis. Finally, the phrase structure rules of both the source and target languages could be used to generate a 24 meaningful subset of all possible sentence (i.e. surface) structures for each of the two languages. The constituents of these two subsets of surface structures would need to be mapped from the source to the target language. This mapping would inform the machine translation program that, for example, a sentence consisting of the two constituents NP VP in the source language might be rendered as VP NP in the target language. Once all the pieces described above are in place, machine translation of the text might become more of a mechanical procedure. Government And Binding .K: Government And Binding In this section an analysis will be done using the Government and Binding theory of grammar as it is presented in A Survey of Linguistic Theories by Jerold A. Edmondson and Donald A. Burquest (1991:99-119). Portions of five verses from Matthew chapter 26 were chosen for analysis. Each of these texts was selected because it contained a Greek conjunction in postpositive position. This should yield a moderately interesting analysis which will be adequate to illustrate the Government and Binding approach. For the reader who is not familiar with Koine Greek an explanation of the term "postpositive position" is in 25 order. Greek is a highly inflected language so that for the most part Greek indicates grammatical function with affixes rather than word order. However, the function of word order as a grammatical indicator is not entirely absent. In the case of some Greek conjunctions such as de 'but' its position in a clause is rigidly specified. That position is nearly always the second (and in rare instances the third) word in a clause. Linguists sometimes call words like these which must always occur in a fixed location relative to some other sentence component clitics (Schachter 1985:53). Greek Data With Analysis .K: Greek Data With Analysis .M:1 10) ergon g a r kal o n hrgasato work for a good she did N-Acc Sub Adj-Acc V-Aor-Mid Neu-S Conj Neu-S Ind-3S [NP1] SubConj [NP1] V eij eme to me Prep Pro Acc Acc-1S [ PP ] For she has done a good thing to me. 26 11) pantote g a r to u j ptwxo u j exete always for the poor you have Adv Sub Def-Art N-Acc V-Pres Conj Acc-Masc-P Masc-P Act-Ind-2P Adv SubConj [ NP ] V meq eautwn with yourselves Prep Pro-Gen Gen Masc-2P [ PP ] For you always have the poor with you. 11b) em e d e ou pantote exete me but not always you have Pro Sub Adv Adv V-Pres Acc-1S Conj Act-Ind-2P Pro SubConj [ VP ] But you don't always have me. 15) oi d e esthsan autw they and weighed to him Pro-Nom Sub V-Aor-Act Pro-Dat Masc-3P Conj Ind-3P Masc-3S Pro SubConj V Pro triakonta arguria thirty silver pieces Adj-Indcl N-Acc Neu-P Neu-P [ NP ] And they weighed him thirty pieces of silver. 28) touto gar estin this for is Dem-Pro Sub V-Pres Nom-Neu-S Conj Act-Ind-3S Pro SubConj V 27 t o aima mou the blood of me Def-Art N-Nom Pro-Gen Nom-Neu-S Neu-S 1S [ NP ] For this is my blood. Summary Of Grammatical Tags .K: Summary Of Grammatical Tags 10) NP1 SubConj NP1 V PP 11) Adv SubConj NP V PP 11b) Pro SubConj VP 15) Pro SubConj V Pro NP 28) Pro SubConj V NP Phrase Structure Rules .K: Phrase Structure Rules S' -> COMP S S -> {Adv SubConj NP TNS VP (PP)} {NP-Word1 SubConj (NP-Word2) TNS VP (NP) (NP) (PP)} NP -> {(Def-Art) (Adj) N (Adj) (Pro-Gen)} {Pro } {Dem-Pro } VP -> (Adv) (Adv) V PP -> Prep Pro 28 Sample Phrase Markers .K: Sample Phrase Markers 10) S' | ------------------------- | | COMP S | | | ------------------------------- | | | | | | | | | VP | NP | TNS | | | | | -------------- | | | | | | | ------------------ | | | | | | | | | PP | | ------ | | | | | | | | | | | | | | | | | | | NP-Word1 | NP-Word2 | | ----------- | | | | | | | | | | | | | | | | | | | | | | | | -Wh N SubConj N | V Prep Pro | | | | | | | | | | | | | | | | | ergon g a r kal o n | hrgasato eij eme e 'work' 'for' 'a good' past 'she did' 'to' 'me' Note: "The symbol e represents an empty node indicating that the -Wh word in English statements has no phonological manifestation" (Edmondson and Burquest 1991:111). 29 11b) S' | | ----------------------------- | | COMP S | | | | | ------------------------------------- | | | | | | | | | | -Wh NP | TNS VP | | | | | | | | | | | NP-Word1 | | ------------------------- | | | | | | | | | | | | | | | Pro SubConj | Adv Adv V | | | | | | | | | | | | | | | em e d e | ou pantote exete e 'me' 'but' present 'not' 'always' 'you have' .M:2 Critique .K: Critique This subsection will compare the Government and Binding (GB) grammatical approach to other grammatical theories and attempt to answer the question, "To what extent would a Government and Binding analysis be useful in implementing a machine translation program?" In the interest of completeness a variety of aspects of the GB model will be discussed whether or not they are evident from the above data which is necessarily limited. 30 When we compare GB to the Aspects-model of Trans formational Grammar, which was discussed in the previous section, we see that the GB model treats the various parts of the grammar as independent units. This modular approach to grammatical analysis would be particularly appealing to grammarians who prefer to limit the scope of their re search. One researcher might prefer, for instance, to concentrate on surface structures. The modular approach also allows a degree of interchangeability so that one might experiment with a variety of transformational compo nents without changing the remainder of the analysis. This last feature would be of particular appeal to someone trying to develop a machine translation program since some experimentation with the transformational component would be a necessity. The transformational rules of GB are weaker and more general than the rules of Tagmemics, Stratificational Grammar, or the Aspects model of Transformational Grammar. This results in more ungrammatical sentences being gener ated by the grammar. However, some ungrammatical sentences are prevented by constraints, which operate in tandem with the rules, and others are removed by filters, which operate after the rules. This general approach lends itself to the development of a universal grammar, which would attempt to capture those linguistic generalizations that hold for all 31 languages. The division of the grammar into rules, constraints, and filters is another example of modularity which would tend to facilitate development of a machine translation program. The GB approach emphasizes general principles rather than specific transformational rules. Some examples are the principle of subjacency and the principle of the tensed S condition. This new emphasis is an improvement over the Aspects model. The general principles apply to more of the world's languages, and specific rules can be developed from the general principles to treat the specific cases found in a particular language. This feature is perhaps of less interest to those who are developing a machine translation program since they will find it necessary to deal for the most part with specific rules rather than general prin ciples. The phrase structures of the GB model use what is called the X-bar syntax. With this type of notation a noun phrase N'' can have as one of its constituents a noun phrase N', which in turn can have as one of its constitu ents a noun N. This provides an unlimited variety of phrase types between the start symbol and the terminal symbols. This approach makes GB still more general. With X-bar syntax it is not necessary to posit all the different 32 phrase types of a language before an analysis can begin. This type of syntax is another aspect of GB which would be particularly appealing to a person developing a machine translation program. Computer programmers and grammarians alike would say that the description of NP is recursive, in that an N'' may have as one of its constituents an N'. The advantage of such recursive definitions to the computer programmer is that the NP need only be defined once with the provision that other NPs may be embedded within any NP. The GB model of grammar makes use of empty cate gories. They are known as: empty, trace, PRO, and COMP. Use of these empty categories makes possible the universal linguistic principle called alpha-movement. For instance, under alpha-movement a NP might be moved to an empty node leaving behind a trace and converting an active construc tion to a passive in the process. The use of empty cate gories by the GB model allows a certain amount of the deep structure to be represented in the X-bar syntax along with the surface structure. This places less of a burden on the transformational component of the grammar, and simplifies the task of sentence generation, since the target locations of any potential moves are marked with empty categories. If sentence generation is simplified in the grammar, the task would be simplified in a machine translation program as well. 33 Turning now to a specific issue in the analysis of Greek syntax, there are significant semantic problems with any treatment of the Greek particle gar 'for', but the par ticular problem studied in the Koine Greek data above is that of the syntactic function of the postpositive subordi nating conjunction. The GB model does not handle this phenomenon particularly well, but neither does any other grammatical model known to this writer. At the heart of the problem is something called level skipping. The post positive conjunction in Greek is required to occur as the second word in its clause regardless of what constituent might otherwise be occupying that position (See verse 10 above). This results in modifications being made to con stituents which are lower in the grammatical hierarchy than the clause (i.e. level skipping). For instance, where an English speaker would say for the man ... the correct Greek word order would be, the for man .... Using Greek word order the constituents of the noun phrase the man are interrupted by a clause level constituent for. In the rules above the decision was made to ignore the postpositive conjunction at levels lower than the clause. This necessitated the invention of a mechanism to specify the first and second words in a phrasal constit uent. The mechanism chosen was Word1 and Word2. 34 In a machine translation program postpositive conjunctions would probably best be handled by a rule of movement which would operate on subordinate clauses after all other rules, constraints, and filters had been applied to the clause. Thus the subordinating conjunction would be moved to the extreme left constituent position of the rule for the subordinate clause (S in the rules above), and an additional rule of movement would be applied after the subordinate clause was fully formed. The ordered rules would be: 1. S -> SubConj Adv NP TNS VP (NP) (NP) (PP) 2. S -> S-Word2 SubConj S-Word3 The necessity of employing the above scheme is not a negative reflection on the GB model of grammar when compared to other models of grammar since none of the others seems to handle postpositive conjunctions in an elegant manner either. On the contrary, inclusion of a mechanism like Word1 and Word2 to the GB model would be a welcome extension. Finally, it must be noted that the GB model does not deal with semantic issues such as the use of metaphor in verse 28, or the use of forefronting to produce emphasis in 35 verse 11b. Semantic issues will be taken up in more detail in the chapter on semantic structure analysis. Montague Grammar .K: Montague Grammar In this section an analysis will be done using Montague grammar as it is presented in A Survey of Linguistic Theories by Jerold A. Edmondson and Donald A. Burquest (1991:121-131). The text to be analyzed consists of portions of five verses taken from Matthew chapter 26. The text was selected with the goal of adequately illus trating the Montague grammar without making the analysis overly complex. Montague grammar uses a unique set of category names and basic expressions. The items listed here as Table 1 were taken from Montague Grammar edited by Barbara Partee (1976:56). 36 .M:1 Table 1.--Montague Categories ==================================================== Montague Category Meaning ---------------------------------------------------- t truth value expression or declarative sentence e entity expression t/e or IV intransitive verb phrase t/IV or T term or noun phrase IV/T or TV transitive verb phrase IV/IV or IAV adverb which modifies an intransitive verb t//e or CN common noun phrase t/t adverb which modifies a sentence IAV/T prepositional phrase IV/t Verb phrase which can take a that-complement IV/IV Verb phrase which can take an infinitive complement .M:1 ---------------------------------------------------- Greek Data With Analysis .K: Greek Data With Analysis 40) erxetai pr o j to u j maqht a j He comes to the disciples V-Pres-M/P Prep Def-Art N-Acc Ind-3S Acc Acc-Masc-P Masc-P t/e [ IAV ] He came to the disciples. 45) o ui o j tou anqrwpou paradidotai the Son of the Man is betrayed Def-Art N-Nom Def-Art N-Gen V-Pres-Pass Nom-Masc-S Masc-S Gen-Masc-S Masc-S Ind-3S [ t/IV ] t/e 37 eij xeiraj amartwlwn into hands of sinners Prep N-Acc Adj-Gen Acc Fem-P Masc-P [ IAV ] The son of man is betrayed into the hands of sinners. 50) epebalon t a j xeiraj ep i t o n Ihsoun they laid on the hands on the Jesus V-Aor-Act Def-Art N-Acc Prep Def-Art N-Acc Ind-3P Acc-Fem-P Fem-P Acc Acc-Masc-S Masc-S IV/T [ t/IV ] [ IAV ] They laid hands on Jesus. 65) o arxiere u j dierrhcen the high priest tore Def-Art N-Nom V-Aor-Act Nom-Masc-S Masc-S Ind-3S [ t/IV ] IV/T t a imatia autou the garments of him Def-Art N-Acc Pro-Gen Acc-Neu-P Neu-P Masc-3S [ t/IV ] The high priest tore his garments. 69) s u hsqa met a Ihsou tou Galilaiou you were with Jesus the Galilaean Pro V-Impf-Act Prep N-Gen Def-Art Adj-Gen Nom-2S Ind-2S Gen Masc-S Gen-Masc-S Masc-S t/IV t/e [ IAV ] You were with Jesus the Galilaean. 38 Summary Of Grammatical Tags .K: Summary Of Grammatical Tags 40) t/e IAV 45) t/IV t/e IAV 50) IV/T t/IV IAV 65) t/IV IV/T t/IV 69) t/IV t/e IAV Sample Phrase Markers .K: Sample Phrase Markers 40) erxetai pr o j to u j maqht a jt | | -------------------------------- | | erxetait/e pr o j to u j maqht a jIAV | ------------------- | | pr o jIAV/T to u j maqht a jt/IV | ------------ | | to u j? maqht a jt//e 65) o arxiere u j dierrhcen t a imatia autout | --------------------------------------- | | | o arxiere u jt/IV dierrhcenIV/T t a imatia autout/IV | | ------------- ---------------------- | | | | | o? arxiere u jt//e t a ? imatiat//e autou? 39 69) s u hsqa met a Ihsou tou Galilaiout | ----------------------------------------- | | | s u t/IV hsqat/e met a Ihsou tou GalilaiouIAV | ------------------------------ | | met a IAV/T Ihsou tou Galilaiout/IV | -------------------------- | | Ihsout/IV tou Galilaiou? | -------------- | | tou? Galilaiou? .M:2 Critique .K: Critique This subsection will compare the Montague gram matical approach to other grammatical approaches and attempt to answer the question, "To what extent would a Montague analysis be useful in implementing a machine translation program?" In the interest of completeness a variety of aspects of the Montague model will be discussed whether or not they are evident from the above data which is necessarily limited. Montague grammar deals largely with the surface structure. It does not assume that there is a unique deep structure level. However, in the case of a syntactic derivation involving the Rule of Quantification it is 40 possible to have an expression appear somewhere other than its final location. For instance, a pronoun form can be replaced by a noun phrase from the context outside the sentence. Therefore, in this sense there seem to be movements and transformations, but these aspects of Montague grammar fall short of the usual surface versus deep structure distinction. Montague grammar as found in Montague's The Proper Treatment of Quantification in Ordinary English (1974:247- 270) has only two primitive symbols. The symbol t stands for a truth-value expression (declarative sentence). The symbol e stands for an entity expression or an individual expression. All the other symbols/categories are built up from these two primitives by mathematical derivations. The t symbol is commonly associated with a sentence, and the e symbol with a noun phrase. However, there is also a derived symbol which can represent a noun phrase. That symbol is t/IV, which is abbreviated T, and called a term. The mathematical derivations of these derived symbols work like fractions in mathematics in the sense that it is possible to reduce these fractions. A thoughtful examination of Table 1 and the sample phrase markers above will reveal some rather troubling 41 omissions. For instance, Montague does not include among the basic or derived categories the category of adjective. In addition, articles (quantifiers) are not introduced into sentences by means of categories. Rather, quantifiers such as every, a, and the are treated as being syncategorimatic, having no category in themselves, but nevertheless appear ing as part of other complex expressions. This idea is a carryover from logical calculi. The categories of adjective and article are indi cated by question marks in the phrase markers. Such categories are fundamental and necessary in describing a language, not to mention making a machine translation to or from one. However, this is neither a criticism of Montague nor of his grammar since he does not claim to have produced an exhaustive description of English. On the contrary, in The Proper Treatment of Quantification in Ordinary English he only claims to deal with "the syntax and semantics of a certain fragment of a certain dialect of English" (1974:247). Nevertheless, a linguist attempting to describe a language with Montague grammar would almost certainly find it necessary to extend the grammar to include these cate gories and possibly others as well. In fact, Richmond H. Thomason in Some Extensions of Montague Grammar expands 42 Montague's categories substantially adding categories for adjective, determiner, and others (1976:79). The rules of Montague grammar do not make reference to grammatical relations such as subject and direct object. References are limited to the primitive symbols and derived categories found in Table 1. This is analogous to the situation found in Chomsky's Aspects-model of Transforma tional Grammar and more modern derivatives of TG such as Government and Binding. However, it should be noted that David Dowty (1981) has proposed extensions to Montague grammar which make reference to grammatical relations. On the one hand, the lack of grammatical relations in Montague grammar can hardly be considered a serious shortcoming inasmuch as significant linguistic analysis has been done with all of the Chomsky family of grammatical models, and these models also lack grammatical relations. On the other hand, inclusion of this additional information is bound to be of some significant aid to a person develop ing a machine translation program. It is hard to imagine mechanically transforming the linguistic representation of one language into another when one is unsure which of two noun phrases is the subject and which is the direct object. Nevertheless, this may not be 43 a serious handicap in that such a determination can often be made in some languages by examining some parameter such as the position of a constituent in its sentence. Still, the fact that such a determination might very often have to be made indicates that, for machine translation purposes, Montague grammar could benefit from being augmented in this area. The rules of Montague grammar are presented in a rigorous mathematical syntax and symbology which is indecipherable to the reader who is unschooled in the mathematics of formal logic. While this is something of an obstacle to the unmotivated, it is actually highly appealing to one who is writing a machine translation program. In Montague Grammar and Machine Translation between Thai and English Kurt Godden says, .M:1 ----L----1----+----2----+----3----+----4----+-T--5----R----6----+----7----+----8 MG [Montague grammar] is well-suited to MT [machine translation] because the mathematical rigor in an MG fragment makes implementing the grammar straightforward and also reveals inadequacies as well as strengths in its tightly-knit structure (1981:7). .M:2 L---+----1----+----2----+----3----+----4----+----5T---+---R6----+----7----+----8 Montague grammar emphasizes semantics as well as syntax. This is a definite improvement over the syntax only models of the Chomsky school. With Montague grammar it is possible to use rules to move from a syntactic derivation (sentence) to its representation in intensional logic (a level of the syntax which is closer to the 44 semantic interpretation) and back again to the syntactic derivation. In fact, the latter syntactic derivation does not have to be in the same language! This is how machine translation is accomplished using Montague grammar; the intensional logic is used as an interlingua. Godden says, .M:1 ----L----1----+----2----+----3----+----4----+-T--5----R----6----+----7----+----8 ... a tensed intensional logic comprises the language through which the semantic interpretation of English is indirectly accomplished, ... the intensional logic can be used as the interlingua to map from one language to another (1981:7-8). .M:2 L---+----1----+----2----+----3----+----4----+----5T---+---R6----+----7----+----8 Another point which can be made about the semantics of Montague grammar is that it follows what is called: .M:1 ----L----1----+----2----+----3----+----4----+-T--5----R----6----+----7----+----8 truth functional semantics. This idea found its most rigorous statement in the work of Polish logician Alfred Tarski, who said that to know what a statement means is to know under what conditions it would be true. (Edmondson and Burquest 1991:122) .M:2 L---+----1----+----2----+----3----+----4----+----5T---+---R6----+----7----+----8 A translator, on the other hand, could take a much more pragmatic approach to meaning. For a translator the meaning of a sentence could simply be its translation in the target language. This may seem like a good example of circular logic inasmuch as it is meaning which a translator is supposed to be translating. However, for the translator it is sufficient to know that La casa es blanca means The house is white, and it is unnecessary to know under what conditions such a statement might be true. In this regard, therefore, Montague grammar is perhaps 45 stronger in its handling of semantics than is required to produce a machine translation. Montague grammar does not purport itself to be a model of English or any other language in the psychological sense. Rather, Montague is accustomed to dealing with the artificial language of mathematical logic. Nevertheless, with some extensions Montague grammar can be used to describe a natural language, and in fact, as Godden's work demonstrates, Montague grammar can be used to produce a successful sentence-level machine translation program. Relational Grammar .K: Relational Grammar In this section an analysis will be done using Relational Grammar as it is presented in A Survey of Linguistic Theories by Jerold A. Edmondson and Donald A. Burquest (1991:121-131). The text to be analyzed consists of portions of five verses taken from Matthew chapter 26. Texts with passive verbs were selected in order to illus trate the promotional feature of Relational Grammar. 46 Greek Data With Analysis .K: Greek Data With Analysis .M:1 3) sunhxqhsan oi arxiereij were assembled the chief priests V-Aor-Pass Def-Art N-Nom Ind-3P Masc-P Masc-P Pred [ Subj ] The chief priests were assembled. 24) o ui o j tou anqrwpou paradidotai the Son the of man is betrayed Def-Art N-Nom Def-Art N-Gen V-Pres-Pass Nom-Masc-S Masc-S Gen-Masc-S Masc-S Ind-3S [ Subj ] Pred The Son of man is betrayed. 31A) umeij skandalisqhsesqe en emo i you will be offended in Me Pro-Nom V-Fut-Pass Prep Pro-Dat 2P Ind-2P Dat 1S Subj Pred [ Cho ] You will be offended in Me. 31B) diaskorpisqhsontai t a probata will be scattered the sheep V-Fut-Pass Def-Art N-Nom Ind-3P Nom-Neu-P Neu-P Pred [ Subj ] The sheep will be scattered. 54) plhrwqwsin ai grafa i can be fulfilled the scriptures V-Aor-Pass Def-Art N-Nom Subj-3P Nom-Fem-P Fem-P Pred [ Subj ] The scriptures can be fulfilled. 47 Summary Of Grammatical Tags .K: Summary Of Grammatical Tags 3) Pred Subj 24) Subj Pred 31A) Subj Pred Cho 31B) Pred Subj 54) Pred Subj Sample Stratal Diagrams .K: Sample Stratal Diagrams ___________________________ 3) | | P | 2 | -----|-------------------------|----- | | P | 1 | -----|-------------------------|----- | | | | sunhxqhsan oi arxiereij were assembled the chief priests ___________________________ 24) | | 2 | P | -----|-------------------------|----- | | 1 | P | -----|-------------------------|----- | | | | o ui o j tou anqrwpou paradidotai the Son the of man is betrayed 48 ____________________________________ 31A) | | | 2 | P| 1 | -----|---------------|------------------|----- | | | 1 | P| Cho | -----|---------------|------------------|----- | | | | | | umeij skandalisqhsesqe en emo i you will be offended in Me .M:2 Critique .K: Critique This subsection will compare and contrast Relational Grammar with other grammatical theories. In the interest of completeness a variety of aspects of Relational Grammar will be discussed regardless of whether they are evident from the data above. This section will also attempt to answer the question, "To what degree would a Relational analysis be useful in implementing a machine translation program?" Relational Grammar, as its name implies, concerns itself with grammatical relations. It does not deal with lower level phenomena such as phonology or morphology, nor does it concern itself with semantics or the analysis of discourse. The relations which this grammar deals with are: the predicate, the subject, the direct object, the indirect object, and chomeur which is a French word for 49 'unemployed'. In the stratal diagrams of Relational Grammar these relations are abbreviated: P, 1, 2, 3, and Cho respectively. Relational Grammar does not distinguish between the oblique relations such as location, path, instrument, and possessor; these relations as well as those of predicate and chomeur are considered to be non-terms in Relational Grammar. Non-terms are of little interest because they seem to have little bearing on changes of grammatical relations. This is not to say that non-terms are linguistically unimportant, but only that their functions are not in focus in Relational Grammar. The grammatical relations: subject, direct object, and indirect object are numbered because the subject grammatical relation is considered to be of higher rank than either the direct object or the indirect object. Similarly, the direct object is of higher rank than the indirect object. The chomeur grammatical relation has the lowest possible rank, while the predicate grammatical relation is not given a rank since its rank does not change. Of primary interest to the relational grammarian are promotions and demotions of grammatical constituents to relations of higher or lower order. Such a promotion can be seen in the English passive where a direct object 50 constituent of a sentence is promoted to the subject relation. From the perspective of Relational Grammar this is what happens in the generation of a passive sentence such as: .M:1 Active 1 P 2 John hit Bill. Passive 1 P Cho Bill was hit by John. .M:2 From the point of view of Relational Grammar the active form of the sentence in the example above is con sidered to be of a higher stratum. The passive form of the sentence is considered to be of a lower stratum--having been derived from the form in the higher stratal level by the mechanism of 2 to 1 advancement. Thus Relational Grammar "assumes that various syntactic strata exist in the clause" (Edmondson and Burquest 1991:167). These syntactic strata are illustrated by the stratal diagrams given above. In the diagrams the strata are separated by horizontal lines, and the grammatical relations for each stratal level are indicated above those lines. The earliest stratum is at the top of each diagram, and the final stratum is at the bottom. The syntactic 51 constituents corresponding to each grammatical relation are found at the bottoms of the vertical lines. In the final level of the stratal diagram for the Greek verse (31A) above the reader will find the Greek equivalent of the following passive sentence: .M:1 1 P Cho (31A) You will be offended in Me. .M:2 Notice that in the next higher stratal level in Me has the subject or 1 grammatical relation, and You has the direct object or 2 relation. From the point of view of Relational Grammar the active form of the sentence has undergone a 2 to 1 promotion creating the passive form of the sentence. Since the former subject of the sentence can no longer occupy the 1 position, it moves to chomeur. In English such a transformation from the active form of a sentence to its passive form would be indicated by word order. However, in Greek word order is used primarily to place emphasis on certain constituents, and has no bearing on the active versus passive status of a sentence. In Greek, changes in case markings serve the same function that word order does in English. Thus (31A) en emoi 'in Me' is expressed in one of the oblique cases (dative case) while its grammatical relation is chomeur. 52 However, were the sentence in its active form, en emoi could not be used; some form in the nominative case could be used, or the subject could simply be expressed via verb conjugation. Also, the subject umeij 'you' would have to be expressed in the accusative, and the verb form used in the active would have to be different since Greek has different forms for active and passive verbs. Regardless of the lexical selections made in the process, the basic word order could remain unchanged. Other languages allow both promotion (passives) and demotion (antipassives and inversions). In the antipassive the subject of a sentence is said to retreat to the direct object relationship. This retreat forces the original direct object into a chomeur relationship. Finally, the direct object is said to re-advance to the subject position by virtue of the Final-1 Law of Relational Grammar. The Final-1 Law says that there must be a 1 (subject) in the final stratum of a sentence's stratal diagram. For examples of the antipassive in Choctaw and Eskimo see Edmondson and Burquest (1991:159-160). The rules of Relational Grammar do not produce a constituent structure such as those produced in Trans formational Grammar. Only networks of grammatical rela tions are produced. However, these grammatical relations 53 are themselves lacking in Transformational Grammar. There fore, it would seem that the two systems of grammatical analysis are to some degree complementary--Transformational Grammar being strong in dealing with constituents, and Relational Grammar strong in dealing with grammatical relations. Next, it should also be pointed out that Relational Grammar lacks the generative component that Transformational Grammar has. Finally, it should be noted that Relational Grammar concerns itself more with universal principles than with the specifics of any given language. For purposes of producing a machine translation program Relational Grammar has something to offer that Transformational Grammar does not--namely grammatical relations. However, Relational Grammar lacks a semantic component as well as generative and transformational components, both of which would be of great usefulness in the construction of a machine translation program. Then again, even Transformational Grammar lacks transformations from one language to another, but it would seem that in order to construct rules for such transformations it would be necessary to take grammatical relations into account. Therefore, both Relational Grammar and Transformational Grammar have things to offer in the development of a machine translation program, but even when these aspects of the two grammars are combined, the semantic component 54 and the necessary language-to-language transformational component are still lacking. Role And Reference Grammar .K: Role And Reference Grammar In this section an analysis will be done using Role and Reference Grammar (RRG) as it is presented in A Survey of Linguistic Theories by Jerold A. Edmondson and Donald A. Burquest (1991:146-152). The text to be analyzed consists of portions of five verses taken from Matthew chapter 26. The texts were selected with the goal of clarity and simplicity of the analysis in mind. Greek Data With Analysis .K: Greek Data With Analysis .M:1 55) kaq hmeran en tw ierw ekaqezomhn With day in the temple I sat Prep N-Acc Prep Def-Art Adj-Dat V-Impf-M/P Acc Fem-S Dat Dat-Neu-S Neu-S Ind-1S [ Prep Asgn ] [ Prep Asgn ] [Act/Pred ] Every day I sat in the temple. 58) o Petroj hkolouqei autw ap o makroqen The Peter followed Him from afar Def-Art N-Nom V-Impf Pro-Dat Prep Adv Nom-Masc-S Masc-S Act-Ind-3S Masc-3S Gen [ Act ] [ Pred ] [ Und ] [ Prep Asgn ] Peter followed Him from afar. 55 61) dunamai katalusai I am able to destroy V-Pres V-Aor-Act M/P-Ind-1S Inf [ Act/Pred ] t o n na o n tou qeou the shrine of the God Def-Art N-Acc Def-Art N-Gen Acc-Masc-S Masc-S Gen-Masc-S Masc-S [ Und ] I am able to destroy the temple of God. 64) ap arti oyesqe from now you will see Prep Adv V-Fut-Mid Gen Ind-2P [Prep Asgn ] [ Act/Pred ] t o n ui o n tou anqrwpou the Son of the Man Def-Art N-Acc Def-Art N-Gen Acc-Masc-S Masc-S Gen-Masc-S Masc-S [ Und ] From now on you will see the Son of Man. 65) o arxiere u j dierrhcen the high priest tore Def-Art N-Nom V-Aor-Act Nom-Masc-S Masc-S Ind-3S [ Act ] [ Pred ] t a imatia autou the garments of him Def-Art N-Acc Pro-Gen Acc-Neu-P Neu-P Masc-3S [ Und ] The high priest tore his garments. 56 Summary Of Grammatical Tags .K: Summary Of Grammatical Tags 55) Prep Asgn Prep Asgn Act/Pred 58) Act Pred Und Prep Asgn 61) Act/Pred Und 64) Prep Asgn Act/Pred Und 65) Act Pred Und Sample Clause Mapping .K: Sample Clause Mapping 58) Synthetic Structure op[c[o Petroj n[hkolouqei]n autw]c ap o makroqen]op The Peter followed Him from afar | | | | | | | | Syntactic | | | | Status: | | | | | | | | | | | | Act Pred Und Prep Asgn | | | | | | | | Semantic | | | | Roles: | | | | | | | | | | | | Ag Pred Go Loc | | | | | | | | Logical | | | | Structure: | | | | | | | | ----- ------ ------- --------- | | | | | | | | | | | | [DO (Peter, [do (Peter, follow' (Peter, Him))])] [from afar] .M:2 57 Distinctives Of Role And Reference Grammar .K: Distinctives Of Role And Reference Grammar This subsection will compare and contrast Role and Reference Grammar with other grammatical theories. In the interest of completeness a variety of aspects of Role and Reference Grammar will be discussed regardless of whether they are evident from the data above. This section will also attempt to answer the question, "To what degree would a Role and Reference analysis be useful in implementing a machine translation program?" Role and Reference Grammar, as its name implies, concerns itself with "the interaction of role (semantic) and referential (pragmatic) factors in grammatical systems" (Foley and Van Valin 1984:16). This is a decidedly broader view of grammar than that taken by the generative trans formational family of grammatical theories. Role and Reference Grammar distinguishes itself from the generative transformational tradition in that RRG takes a more func tional approach to grammar rather than a strictly formal approach. The formal approach to grammar views language as structure. Thus from a formalist's point of view a lan guage has been described when its structure has been described. The formalist's structural description of a 58 language would include: (1) a description of its phonology, (2) a description of its morphology, (3) a description of its syntax, and (4) a restricted definition of the meaning of a sentence as derived from its constituents and their configuration. Foley and Van Valin say that when a struc tural approach to language is followed "'language', as a general concept, is in effect reduced to 'grammar', and accordingly linguistics, the study of language, is reduced to the study of grammar" (1984:3). Linguists adhering to a functional approach to grammar believe that "language must be studied in relation to its role in human communication" (Foley and Van Valin 1984:7). The role which language plays in human commu nication is what is referred to above as the referential or pragmatic factor of grammatical systems. "Language is thus viewed as a system of human communication, rather than as an infinite set of structural descriptions of sentences" (Foley and Van Valin 1984:7). Furthermore, Foley and Van Valin contend that "an understanding of language structure requires an understanding of the functions language can serve, communication being the primary one" (1984:9). Functionalists will also point out that there is more to linguistic competence than just grammatical compe tence. Linguistic competence is more than just the ability 59 to generate correct sentences; groups of sentences must be correctly generated as well. Speakers of a language have a sense of whether or not a series of sentences properly conveys concepts or relationships relevant to the context. If the series communicates concepts which seem logical and reasonable in a given context then it is said to constitute a text. Thus, a part of linguistic competence is textual or discourse competence. Such competence "explicitly involves both linguistic and social knowledge" (Foley and Van Valin 1984:11). Additionally, functionalists take an entirely different view of linguistic universals. A formalist looks for universals among the rules and principles which govern the generation of sentences in the world's languages. A functionalist, on the other hand, looks at entire systems of human communication, and for each system asks how each whole system is the same or different. The functionalist is looking for more than just morphosyntactic universals. The functionalist is also looking for discourse universals and sociolinguistic universals. Foley and Van Valin say that discourse and sociolinguistic universals interact with morphosyntactic universals "yielding universals concerning the interplay of form and (communicative) function in human language" (1984:12). 60 Finally, Foley and Van Valin say that a function alist views a system of communication as consisting of: .M:1 ----L----1----+----2----+----3----+----4----+-T--5----R----6----+----7----+----8 (1) a system of verbal semantics and argument functions; (2) a morphosyntactic system; (3) a pragmatic system including notions such as illocutionary force, presupposition, topicality, and definiteness; and (4) a system of social norms governing different kinds of speech events and activities (1984:14). .M:2 L---+----1----+----2----+----3----+----4----+----5T---+---R6----+----7----+----8 A functionalist is concerned with the components of each of these four systems and how they contrast, combine, and interact. Description Of Role And Reference Grammar .K: Description Of Role And Reference Grammar Role and Reference Grammar proposes only two levels of grammatical analysis: "a semantic 'logical structure' in which the predicate of a clause and its arguments are represented, and the actual morphosyntactic form of the utterance" (Foley and Van Valin 1984:15). At the syntactic level two relations are posited: actor and undergoer. These syntactic relations map into semantic roles such as agent and patient at the semantic level. In fact there is a hierarchy of such semantic roles such that the syntactic relation of actor is most likely to be realized first as agent, then as effector, then as locative, and so on. The 61 hierarchy is simply reversed for the syntactic relation of undergoer (Edmondson and Burquest 1991:147). Role and Reference Grammar also employs a scheme of lexical decomposition which provides a mechanism for classifying verbs. This scheme is implemented by means of: "semantic predicates, operators, and connectives" (Edmondson and Burquest 1991:148). With this scheme (which is similar to Stratificational Grammar but with fewer strata) it is possible to represent diagrammatically the various semantic interpretations of a sentence. These semantic interpretations can then be mapped back into their morphosyntactic or surface forms. Such a mapping provides a visual representation of the difference in meaning between sentences which are only subtly different, such as the active and passive realizations of a single semantic interpretation. Role and Reference Grammar divides the morpho syntactic representation of a sentence into several layers: (1) the nucleus which is the predicate, (2) the core which consists of the arguments of the predicate, and (3) the inner and outer periphery which is defined as any other arguments which are optional but not required by the predicate. The inner periphery consists of those periph 62 eral arguments which are coded in the verb; the outer periphery encompasses any other peripheral arguments. Discussion Of The Analysis .K: Discussion Of The Analysis The sample clause mapping following the Greek data above will serve to illustrate the analytical technique of Role and Reference Grammar. At the top of the mapping is the Greek text for the portion of verse (58) which is being analyzed. The text is divided into three sections using square brackets. The innermost section is the nucleus. Surrounding the nucleus is the core, and the outermost division is the outer periphery. The nucleus contains the predicate. The core contains arguments of the predicate; i.e. subject, direct object, and indirect object. The periphery contains other arguments of the clause; i.e. prepositional phrases. Inner and outer peripheries are distinguished by whether or not their arguments are tied to the predicate. The syntactic status of each constituent is indi cated below the clause. Of particular interest in Role and Reference Grammar are the syntactic relations of actor and undergoer. These syntactic relations are mapped into a number of semantic roles. In this case the actor is mapped 63 into the semantic role of agent, and the undergoer is mapped into the semantic role of goal. The constituent identified syntactically as prepositional assignment maps into the semantic role of locative. This last constituent is of less interest because it is optional and not repre sented in the logical structure of the predicate. Beneath the semantic roles is the logical structure of the clause. The predicate follow' is marked with an apostrophe to indicate that it is a stative verb. (In Role and Reference Grammar all predicates are represented as stative verbs. An exception to this rule is the group of predicates which also function as abstract operators; i.e. DO, BECOME, and CAUSE.) The predicate follow' is transitive and thus takes a subject Peter and a direct object Him as its arguments. The function of the innermost do (abstract operator/predicate) is to transform the predicate follow' from a state to an activity. (The predicate do takes two arguments: an agent, and a predicate which may have arguments of its own.) This innermost do has as its arguments Peter (agent) and follow' (predicate) with its arguments. The outermost DO has as its arguments Peter (agent) and the abstract operator/predicate do with its arguments. This outermost DO serves to transform the core of the logical structure from a non-volitional to a volitional activity. (It is customary to write the DO 64 which performs this function with capital letters.) The locative, as part of the outer periphery, is an optional clause constituent, and is not represented in the logical structure of the verb. For this reason it is isolated from the core. Relevance To Machine Translation .K: Relevance To Machine Translation Role and Reference Grammar lacks grammatical rela tions; i.e. subject, direct object, etc. As a practical matter, a machine translation program needs to keep track of grammatical relations because these are what determine such things as inflections for case (if any) in the surface representation of the translated text. However, this does not seem to be a serious drawback in that there is nothing inherent in Role and Reference Grammar which would prohibit the incorporation of grammatical relations into the scheme. Role and Reference Grammar does not concern itself with constituency. This is not to imply that phrasal constituents are unimportant, but only that RRG does not focus on these. While a machine translation program must concern itself with all levels of constituents, there seems to be no reason why this level of analysis could not be incorporated into an augmented Role and Reference Grammar. 65 Role and Reference Grammar does concern itself with the logical structure of of a clause. Role and Reference Grammar's definition of a clause's logical structure goes a long way towards being a semantic representation of the clause. Such a semantic representation is the deep structure or interlingua that serves as a machine trans lation program's bridge between languages. However, the formal description of RRG's logical structure is incomplete at present. For instance, the arguments of only a handful of predicates have been defined, and there is no formal mechanism tying the outer periphery to the core. Further more, the mappings of surface level constituents to seman tic level constituents is only represented schematically; there are no formal rules representing these relationships. However, there seem to be no flaws in the RRG definition of logical structure to the extent that it has been developed. For this reason it is plausible that it could be further developed and augmented to serve as a very practical component of a machine translation program. In summary, Role and Reference Grammar lacks some of the features which would make it a practical tool for use in developing a machine translation program. However, none of these deficiencies are inherent. Furthermore, RRG's definition of the logical structure of a clause, while not fully developed, makes it a very attractive prospect for 66 further development and subsequent use in developing a machine translation program. Conclusion .K: Conclusion In this chapter five different approaches to sen tence level grammatical analysis have been explored and evaluated. Each was found to have something to contribute to the successful implementation of a machine translation program. Yet, none was found to be completely sufficient in this regard. In the next chapter a method of analyzing spans of text larger than a sentence (Semantic Structure Analysis) will be examined and evaluated with regard to its potential contribution to the implementation of a machine translation program. In the final chapter an actual machine translation program is described. It will be apparent at that point that features from each of the five sentence level approaches to grammatical analysis as well as features from a Semantic Structure Analysis have been incorporated into the program.