Sintaksinė žodžių trauka

DAIVA ŠVEIKAUSKIENĖ Institute of the Lithuanian Language Research areas: computational linguistics, automatic syntactic analysis. SYNTACTIC PULL OF WORDS Syntactic Attraction of Words A B S T R A C T During computer-based syntactic analysis of Lithuanian, a new phenomenon was observed that had not yet been described in the literature. If a sentence contains two words that govern the same case, and that case occurs between them, which word will form a phrase with the governed case, and which will remain without an expansion? The syntactic pull of participles, nouns, and infinitives is described, and the observed cases are compared with their equivalents in other languages. The term syntactic pull of words itself is also defined. A B S T R A C T By creating syntactic analysis of the Lithuanian language revealed a new phenomenon, which has not been previously discussed in literature. If there are two words in a sentence governing the same case and that case stands in-between them, which of the words will form a word phrase with the governed case and which of them will remain unexpanded? The article describes the syntactic pull of the participle, the noun and the infinitive; comparisons of the cases identified and the equivalents from other languages are provided. The term of syntactic pull of words is defined as well. 1. INTRODUCTION The advent of computers made it possible to automate the intellectual work performed by humans. From 1968 to 1978, the grammatical analysis of natural languages was the main topic for artificial intelligence researchers (Nirenburg 1987: 26). However, natural languages, K E Y W O R D S: automatic syntactic analysis; syntactic structure of a sentence; syntactic relations; syntactic pull of words; syntactic pull of participle, infinitive and noun. K E Y W O R D S: automatic syntactic analysis; syntactic structure of a sentence; syntactic relations; syntactic pull of participle, infinitive and noun. 130 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka which are easily mastered by humans, proved difficult for computers to handle (Jensen, Heidorn, Richardson, 1993: 2). Therefore, when natural languages are processed automatically, results as good as those achieved in other areas of computer use have not yet been attained. 1.1. Work on Computerizing Syntax in Lithuania Perhaps the first attempt to describe a method of automatic syntactic analysis used for Lithuanian can be found in Vidas Daudaravičius’s master’s thesis, whose main aim was “to establish the basic principles of automatic syntactic analysis of the Lithuanian language (The main goal of this research is to set general principles for automatic syntactical analysis of Lithuanian language)” (Daudaravičius 2002: 6). A graph of relations is presented, representing knowledge about a particular language (in this case, Lithuanian), that is, syntactic relations between words. The nodes of the graph represent grammatical categories, while the arcs indicate relations between those grammatical categories. The graph of relations (Fig. 1) (Daudaravičius 2002: 38) was constructed on the basis of Vitas Labutis’s syntax (1996). 1 PAV. Lietuvių kalbos ryšių grafas (Daudaravičius 2002: 38) Straipsniai / Articles 131 DAIVA ŠVEIKAUSKIENĖ From this graph, a dependency tree is constructed for each sentence, encompassing all possible variants if the sentence is ambiguous. Tree formation is divided into two stages: first, a graph of relations is drawn, representing all possible syntactic relations in a particular sentence; then a dependency tree is extracted from it (Daudaravičius 2002: 55). The analysis is illustrated with the sentence Graži mergaitė stebi gražią gėlę pievoje. According to the morphological analyzer’s data (Fig. 2), a graph of syntactic relations is first formed (Fig. 3), showing all possible relations between the words in the sentence; a dependency tree is then derived from it—that is, the syntactic structure of the sentence, or several trees if the sentence is syntactically ambiguous. 2 PAV. Morfologinio analizatoriaus duomenys (Daudaravičius 2002: 35) FIG. 3. All possible word combinations in the sentence Graži mergaitė stebi gražią gėlę pievoje (Daudaravičius 2002: 35) The sentence presented has five possible variants, so all of them are shown in the dependency graph (Fig. 4) and each is presented separately as an equally valid option (Fig. 5). 132 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka FIG. 4. All possible variants of the sentence A pretty girl watches a pretty flower in a meadow (Daudaravičius 2002: 46) FIG. 5. All possible variants of the sentence A pretty girl watches a pretty flower in a meadow (Daudaravičius 2002: 47) That was where the work begun by the author also came to an end, since he defended his dissertation on a different topic and noted there: “It is naive to think that methods successfully applied to English are suitable for other languages as well” (Daudaravičius 2012: 3). Straipsniai / Articles 133 DAIVA ŠVEIKAUSKIENĖ Another master's thesis on computerizing syntactic analysis was written by Gintarė Grigonytė. She uses the unification (compatibility) method, which is described extensively by Peter Hellwig (Hellwig 2003). Generalized rules are formulated, for example, that two elements must agree in number and case. “The system evaluates which rules can be applied to which words in the sentence. In this way, a structure is created from the words in the sentence and the rules that connect them. After all available rules have been examined, a list of the sentence's word combinations is formed” (Grigonytė 2006: 29). Trees are formed from this list of word combinations by applying a fitness function. The fitness function is calculated on the basis of syntactic rules and information about the sentence. The input data for the analysis system is a Lithuanian sentence, and the result is an abstract representation of it. The result of processing the sentence The sun was slanting westward is shown in Fig. 6. 6 PAV. Sakinio Saulėlė krypo vakarop abstraktus pavaizdavimas (Grigonytė 2006: 40) The thesis notes that “The implemented system occupies an intermediate position on the system life cycle scale—between a research prototype and a fully functional prototype” (Grigonytė 2006: 36). On the grounds that “The implemented system does not provide a single syntactic analysis solution in every case,” the system's performance was evaluated not at the sentence level but at the level of sentence parts, i.e., by assessing how accurately the system identifies sentence parts; no assessment was made of how accurately it constructs a tree of the sentence's syntactic structure. To test the system, 20 Lithuanian sentences were analyzed. One of the planned future tasks mentioned is “Expand the system and adapt it for treebanks (English tree 134 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka banks) to be created.” However, this plan was never even begun, since the author, like Vidas Daudaravičius, defended her dissertation on a different topic. The work she had planned—to syntactically annotate a Lithuanian corpus—was begun by entirely different researchers in 2013 at the Institute of the Lithuanian Language (Internet link 1). 1.2. Lack of linguistic research Computers should perform better than humans in every field. The only problem is that we are unable to tell them clearly enough how to carry out a given task. Some things become apparent only when we begin to computerize language. What is self-evident to us, learned in childhood without wondering why it is so, has to be stated formally to a computer, i.e., in a language it can understand. This sometimes makes additional research necessary—research that is irrelevant from a linguistic perspective and therefore had not been conducted before. As early as 1969, in a discussion of machine translation (syntactic analysis is part of machine translation—its second stage), it was stated: “The needs of machine translation in fact compel linguists to deepen their theoretical knowledge in various areas of language that have hitherto been insufficiently investigated” (Ivič 1973: 256). 1.2.1. Identifying sentence constituents Before the advent of computational linguistics, linguists studied sentences and syntax as the result of structural formation. The finished sentence structure was examined. There was no interest in how and on what basis a person decides that a particular word performs this rather than another syntactic function. In the computer era, the process of forming a structure becomes the most important matter and the principal object of scientific research. Therefore, sometimes one must look at a sentence unconventionally, that is, somewhat unusually, not as it was viewed before. Grammars generally describe a sentence constituent (for example, the subject) by listing the morphological forms that can perform this function (Ambrazas 1997: 487). That a word is a subject is taken as an axiom, and only the parts of speech that can serve as one are discussed. In earlier grammars, the mechanism for identifying the subject is described in greater detail by Tuputis. Vytautas Sirtautas attempts to analyze problematic cases as well, for example, when a sentence contains two words (nominatives) that could serve as the subject: “When a sentence has a subject, the second nominative usually can no longer occupy the subject position. (He has been ill for the third day already), but performs other syntactic functions” (Sirtautas 1978: 35). Here, that the first nominative is the subject is taken as an axiom (without even attempting to identify it and determine whether the first nominative really performs the subject function), and only then is the question of the second nominative considered. Computational sentence analysis has shown Straipsniai / Articles 135 DAIVA ŠVEIKAUSKIENĖ the falsity of the cited claim, which had not occurred to people until now: in the sentence He has been ill for the third day already, the second nominative serves as the subject. A fundamentally new method is needed here for identifying sentence constituents (in this case, the subject). It is necessary to consider not only which part of speech can express the subject and what morphological forms it may take, but, most importantly, how to determine which word is the subject. Of two nominatives in a sentence, the one that serves as the subject is not the one that comes first or second, but the one that is a personal pronoun. And this was noticed only in the course of computerized language processing. Thus, we must agree with Professor Laimutis Telksnys’s view that our knowledge of the Lithuanian language does not allow Lithuanian syntax to be fully formalized, and that further theoretical research into Lithuanian is therefore necessary. 1.2.2. Identifying word combinations The main characteristic describing the combinatorial possibilities of words is their valency. In Lithuanian, the valency of verbs has been described most extensively. “The meaning of a verb determines certain combinatorial potentialities, which are realized in a sentence when the corresponding words and their forms attach to the verb” (Sližienė 1994: 16). Words that form combinations with a verb are divided into obligatory dependents, which cannot be omitted for a sentence to be grammatically correct, and optional dependents, which may be left unexpressed while the sentence remains grammatical (Sližienė 1994: 32). Words that are not governed by the meaning of the verb can also combine with it. They are called free elements; in this case, the connection with the verb is determined by the word’s capacity to take other words and forms as dependents (Sližienė 1994: 35). This had also been written about earlier: “Words attached adverbially are usually placed next to the words they modify: There I saw a birch that had sprouted l a s t y e a r” (Sirtautas 1978: 21). However, research by Lithuanian linguists stops there. They do not investigate how to determine the dependency of the word last year: whether it is saw last year or last year sprouted. This article is devoted precisely to such investigations. In describing adverbials, it is stated that their inflected components are attached to the predicate and agree with the subject (Jonas returned and saw. She bought it while returning.). It is also emphasized that “attachment is primary here, while agreement is a secondary phenomenon,” that is, adverbials “are more closely associated with the predicate” (Sirtautas 1978: 47). But this is as far as the discussion goes at this level; cases are not analyzed in which an adverbial can attach to two words in a sentence (In those days, the only encyclopedia available to everyone was one published in Moscow. The locative adverbial in Moscow can combine with both available and published) – which of them, then, forms a syntactic relation with the adverbial? It should be added that valency cannot solve the problem described in Section 1.2.1 either. It can only indicate that the word is ill is bivalent and requires, as its complements, a subject in the nominative and an optional object in the instrumental (Sližienė 136 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka 2004: 36). However, it does not indicate how to identify the subject when a sentence contains two nominatives. Therefore, new linguistic research is needed to answer similar questions. This article should be the beginning of precisely such a study. 2. COMPUTER-BASED SYNTACTIC ANALYSIS To clarify how syntactic structure is identified by computer, a detailed sentence analysis will be presented, showing all the steps involved. Fig. 7 shows what information the computer receives from the user, i.e., the sentence is entered as a sequence of characters. First, it is divided into segments—sequences of letters and numbers separated by a space or punctuation mark (Fig. 8). Punctuation marks themselves also form a separate segment; only spaces are ignored: they serve to determine segment boundaries, and that is where their role ends. The sentence is then morphologically analyzed—each segment is assigned the morphological categories of the word used (Fig. 9). After that, some words are assigned semantic features (Fig. 10). It should be noted that this is not a complete semantic analysis and that only features serving the purposes of syntax are used, i.e., those that can determine a word’s syntactic function. For this reason, verb government is also considered a semantic feature here, although in Lithuanian linguistics it is generally classified as syntax rather than semantics. FIG. 7. First step—the sentence received from the user as a sequence of characters. FIG. 8. Second step—the sentence is divided into segments. FIG. 9. Third step—morphological analysis of the sentence is performed. Straipsniai / Articles 137 DAIVA ŠVEIKAUSKIENĖ FIG. 10. Fourth step—some words are assigned semantic features. The following is checked: IS THERE A PERSONAL PRONOUN IN THE NOMINATIVE IN THE SENTENCE; IF NOT, IS THERE A NOUN IN THE NOMINATIVE? This sentence contains a noun in the nominative, and it is assigned the function of subject (Fig. 11). FIG. 11. Fifth step—the subject of the sentence is identified. The next question is: IS THERE A VERB IN THE SENTENCE WHOSE PERSON AGREES WITH THE PERSON OF THE WORD PERFORMING THE SUBJECT FUNCTION? A noun in the nominative corresponds to the third person, and the sentence contains a third-person verb, so it is assigned the function of predicate (Fig. 12). FIG. 12. Sixth step—the predicate of the sentence is identified. 138 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka The sentence analysis continues by asking: IS THERE A CASE GOVERNED BY THE PREDICATE IN THIS SENTENCE? The answer is yes: “Yes, this sentence contains an accusative case governed by the verb functioning as the predicate,” and the noun in the accusative is assigned the function of object (Fig. 13). FIG. 13. Step seven – the object is identified. After identifying the object, the following is checked: IS THERE AN ADJECTIVE, PARTICIPLE, OR NUMERAL OF THE SAME GENDER, NUMBER, AND CASE BEFORE THE NOUN PERFORMING THE OBJECT FUNCTION? This time the answer is no: “No, there are no words that meet these requirements.” The search for adverbials begins next. Thus the computer’s next question is: IS THERE AN ACCUSATIVE WITH A TIME MEANING IN THIS SENTENCE? The answer is: “No.” Then another question is asked: IS THERE A NOUN IN THE LOCATIVE CASE IN THIS SENTENCE? This time the answer is yes, and it is assigned the function of an adverbial of place (Fig. 14). It is also checked whether there are any other types of adverbials, so the question is asked: IS THERE AN ADVERB IN THIS SENTENCE? Yes. It, too, is assigned the function of an adverbial of manner (Fig. 15). FIG. 14. Step eight – the adverbial of place is identified. Straipsniai / Articles 139 DAIVA ŠVEIKAUSKIENĖ FIG. 15. Step nine – the adverbial of manner is identified. After all the words have been processed, punctuation marks are examined: the computer checks: IS THE PERIOD IN THE SENTENCE NOT PART OF AN ABBREVIATION? Having received a negative answer, it assigns the period segment the function of marking the end of the sentence (Fig. 16). FIG. 16. Step ten – the end of the sentence is identified. Finally, a graphical representation of the sentence’s syntactic structure is generated (Fig. 17). Why this particular form is used to analyze syntactic structure is described in greater detail in Šveikauskienė 2005: 407–418. A simplified description of the syntactic analysis process has been presented here, showing how the computer processes this sentence. For simplicity, many operations whose results do not affect the formation of the structure of the sentence under analysis have been omitted. For example, if the sentence contained a modifier of the object (say, interesting), after identifying it, one would need to check whether there is an adverb before the word functioning as the modifier (for example, very) that could function as an adverbial modifying the modifier, and so on. 140 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka FIG. 17. The eleventh step—the syntactic structure of the sentence. Now it is necessary to emphasize what is most important and why this chapter and the article as a whole were written: if we analyze the sentence Tėvas išsamiai aprašė parodą Kaune in the same way, we will obtain the same syntactic structure, which would be an error (in spoken language, using intonation, one can also produce a sentence corresponding to that same structure, but a text contains no information about the intonation with which a sentence was spoken, so questions related to intonation will not be discussed in this article)—the correct structure of this sentence is shown in Fig. 18. However, more data are needed to form it. This is precisely what the following chapters of this article discuss. FIG. 18. The syntactic structure of the sentence Tėvas išsamiai aprašė parodą Kaune. The aim of this chapter was to show why additional information about the conditions under which word combinations occur is needed—that is, why it is necessary to study the ability of words to form a syntactic relation, or, in other words, syntactic word attraction, which is discussed in greater detail in the following chapters of this article. Straipsniai / Articles 141 DAIVA ŠVEIKAUSKIENĖ 3. THE PHENOMENON OF SYNTACTIC WORD ATTRACTION When the syntax of Lithuanian was being formalized, a new linguistic phenomenon was noticed—one that goes unnoticed in everyday use and had not yet been studied by linguists: if a word stands between two other words and can combine with each of them, which of the two will it in fact form a syntactic relation with in that sentence? While developing an automatic syntactic analysis program, and especially while observing the computer's errors during its tuning and testing, it became clear that Lithuanian—and probably other languages with free word order in the sentence—has another characteristic, which I propose to call syntactic word attraction. One reason why the performance of automatic syntactic analysis is poor may be that syntactic word attraction is not taken into account and has not been sufficiently investigated. I will provide several examples in which the computer drew my attention to this phenomenon. It is most noticeable when the word under analysis is placed between two other words that can form a direct syntactic relation with it (the term direct syntactic relation is used in this article in the sense in which it is described in the grammar [Ambrazas 1997: 478]). Then it is easiest to see which of the two words it is attracted to. The sentences used to test the software were taken from the Corpus of Contemporary Lithuanian, compiled at Vytautas Magnus University (Internet link 2). A note that applies to the entire article: only issues related to text are described here, and cases in which intonation can produce other sentence variants in spoken language are not examined. Texts generally contain no data about intonation (unless they are specially annotated texts), so questions related to intonation are left to corpus researchers. 3.1. The syntactic attraction of the participle Perhaps the syntactic attraction of words was first observed in the case of the participle. It is particularly clear that the attraction of a participle functioning as an attribute is stronger than that of a participle functioning as a predicative complement with respect to an adverbial modifier; that is, an attribute expressed by a participle attracts an adverbial modifier (for example, one expressed by a locative) more strongly than a predicate expressed by a participle does. It can also be argued that a participle functioning as an attribute attracts a governed case immediately adjacent to it more strongly than a participle functioning as a predicate does. The predicate attaches a governed case from sentence words that are farther away. If both a participle functioning as an attribute and a participle functioning as a predicate govern the same case, and that case occurs between them, it is the attribute, not the predicate, that forms a phrase with that case. These claims can be illustrated with examples. 142 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka 3.1.1. Adverbial modifier If an adverbial modifier occurs between a predicate expressed by a participle and an attribute expressed by a participle, it modifies the attribute. In sentence (1), the underlined words—a participial attribute and a temporal adverbial modifier—form a direct syntactic relationship. (1) This formula was first mentioned at the meeting held in January between R. Paksas and John Bumgardner, president of “Williams International”. If we remove from this sentence the participle held, which functions as an attribute, then the adverbial modifier will have a direct syntactic relationship with the predicate (as shown in sentence (2)). But if there is an attribute expressed by a participle, the adverbial modifier belongs to it, not to the predicate (even though the predicate is also expressed by a participle), as can be seen in sentence (1). (2) This formula was first mentioned in January at the meeting between R. Paksas and John Bumgardner, president of “Williams International”. Thus, a participle functioning as an attribute attracts an adverbial modifier more strongly than a participle functioning as a predicate does; that is, the attraction of an attribute expressed by a participle is stronger than that of a predicate expressed by a participle with respect to an adverbial modifier. And here it makes no difference what form the predicate takes—whether it is compound or simple. If the compound predicate was first mentioned is replaced by the simple predicate mentioned, the stronger attraction of the participle functioning as an attribute with respect to the adverbial modifier remains; see sentence (3). It contains the participial attribute held, which attaches the adverbial modifier as its own extension. (3) The first time this formula was mentioned at the meeting held in January between R. Paksas and John Bumgardner, president of “Williams International”. In sentence (4), there is no longer a participial attribute held, and the adverbial modifier modifies the predicate expressed by a participle, mentioned, just as when the predicate was compound—was first mentioned. Thus, the form of the predicate does not affect the attraction of an attribute expressed by a participle. (4) The first time this formula was mentioned in January at the meeting between R. Paksas and John Bumgardner, president of “Williams International”. The next example shows that switching the positions of an attribute expressed by a participle and a predicate does not change the syntactic attraction between the words. Thus, regardless of the order in which the participle functioning as an attribute and the participle functioning as a predicate occur in the sentence, if an adverbial modifier lies between them, the attribute expressed by a participle attracts it as its extension. (5) Until then, the only encyclopedia available to many, especially intellectuals, was the one published in Moscow, the Great Soviet Encyclopedia (“Bolshaya sovetskaya entsiklopediya”). (6) Until then, the Great Soviet Encyclopedia (“Bolshaya sovetskaya entsiklopediya”) published in Moscow was the only one available to many, especially intellectuals. Straipsniai / Articles 143 DAIVA ŠVEIKAUSKIENĖ Sentences (5) and (6) merely confirm the claim highlighted in this section: an attribute expressed by a participle allowed attracts an adverbial (i.e., forms a syntactic relation with it) in Moscow; see sentence (5); and only if there is no such attribute (expressed by a participle) (see sentence (6)) does a participle serving as the predicate attach this adverbial accessible. Sentences (5) and (6) also show that a particle inserted between the adverbial and the predicate has no effect on syntactic word attraction. If we switch the predicate (was accessible) and the attribute expressed by a participle (allowed), see sentences (7) and (8), word attraction likewise remains unchanged. As in sentences (5) and (6), the predicate expressed by a participle attaches the adverbial only if the sentence contains no attribute expressed by a participle. (7) Only that allowed in Moscow was accessible to everyone, the “Great Soviet Encyclopedia” (“Bolshaya sovetskaya entsiklopediya”). (8) Only that in Moscow was accessible to everyone, the “Great Soviet Encyclopedia” (“Bolshaya sovetskaya entsiklopediya”). Thus, syntactic word attraction is a fairly independent phenomenon, unrelated to the relative arrangement in the sentence of the words that can attach the adverbial. 3.1.2. Case governed by a word’s meaning If a sentence contains a participle functioning as an attribute and governing the same case as the predicate, it attracts the nearest word in the case it governs. This is particularly evident when a case form governed by both of them stands between the predicate and the participle functioning as an attribute. The attribute expressed by a participle then forms a direct syntactic relation with this word, while the predicate attaches to itself as a complement another word in the same case from further away in the sentence. In sentence (9), both the predicate should submit and the participle functioning as an attribute importing strongly govern the same case—the accusative. (9) Then the entrepreneurs importing them should submit their proposals. Nevertheless, the direct syntactic relation with the adjacent word in the case form governed by both of them (the accusative them) is formed by the attribute expressed by a participle, leaving the more distant (farther-away) accusative proposals to the predicate. Thus, in this case, the attraction exerted by the attribute expressed by a participle is stronger than that of the predicate; in other words, in the immediate context, the attraction of a participle functioning as an attribute is stronger than the predicate’s attraction with respect to the case it governs. 3.1.3. Case governed by the word form Differences in the attraction of a participle and a noun are noticeable in the passive voice. This form governs the genitive, and the participle, functioning as a noun’s agreeing modifier— 144 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka I mean, the genitive preceding it attaches to it as a governed case; that is, a modifier expressed by a passive participle attracts the genitive more strongly than the noun it modifies, see sentence (10). If there is no passive participle, as shown in sentence (11), or an adjective detailed occurs in its place, as in sentence (12), the genitive of the minister remains directly associated with the noun in the report, as can be seen in sentences (11) and (12). (10) Tai teigiama ministro pasirašytame pranešime. (11) Tai teigiama ministro pranešime. (12) Tai teigiama ministro išsam iame pranešime. The adjective of the minister, detailed, which functions as an agreeing modifier, intervenes between the non-agreeing modifier in the report and the word it modifies; this does not disrupt the syntactic relation between the non-agreeing modifier of the minister and the word in the report that it modifies, see sentence (12). However, the attraction of a passive participle is so strong that it is sufficient to break this relation, and it attracts the genitive to itself, making it its complement. Thus, we can say that the attraction of a passive participle is stronger than the attraction of a noun with respect to the genitive. When it stands before the genitive, the passive participle retains its strength of attraction and attaches it (see sentence (13)), whereas an adjective does not (see sentence (14)), analogously to sentences (10) and (12). (13) Tai teigiama pasirašytame ministro pranešime. (14) Tai teigiama išsam iame ministro pranešime. It should be mentioned that in spoken language, intonation can be used to give a sentence a meaning in which the genitive functions as a non-agreeing modifier even when a passive participle is present. However, the text provides no data on intonation (unless the text has been specially annotated), so questions related to intonation are not addressed in this article. 3.2. Noun attraction A noun has a property similar to that of a participle (regardless of whether it functions as a subject or an object): its attraction with respect to an adverbial modifier is stronger than that of the predicate. A couple of examples may be given. The locative case of a noun usually functions as an adverbial modifier of the predicate, but if it immediately follows a noun, it modifies that noun: in sentences (15) and (16), the locative follows the noun (functioning as an object) parodą and modifies it (he could have described the exhibition while being somewhere else, not necessarily in Kaunas). In cases where the locative follows other words, it functions as an adverbial modifier and forms a syntactic relation with the predicate, as can be seen in sentences (17) and (18) (the exhibition itself could also have been somewhere else, Straipsniai / Articles 145 DAIVA ŠVEIKAUSKIENĖ not necessarily in Kaunas). Thus, the attraction of an object expressed by a noun with respect to a locative adverbial modifier is stronger than the attraction of the predicate when the object is in postposition in the locative case. (15) The exhibition in Kaunas he described in detail. (16) He described the exhibition in Kaunas in detail. (17) He in Kaunas described the exhibition in detail. (18) In Kaunas, the exhibition he described in detail described. Another example can be given with a different type of adverbial modifier: for instance, a causal adverbial modifier expressed by a prepositional construction confirms that a noun—in this case functioning as a subject—with such a modifier in postposition exerts a stronger attraction than the predicate. In sentence (19), the prepositional construction due to disrupted transport immediately follows the subject expressed by a noun and modifies precisely that subject, whereas in sentence (20) it functions as a causal adverbial modifier and forms a direct syntactic relation with the predicate. (19) Losses due to disrupted exports through the Klaipėda terminal reached 28 million litas. (20) Due to disrupted exports through the Klaipėda terminal, losses reached 28 million litas. Thus, it can be concluded that when an adverbial modifier occurs in a sentence immediately after a noun functioning as the subject, it is attracted more strongly by that noun than by the verb functioning as the predicate; that is, a noun attracts an adverbial modifier immediately following it more strongly than a verb functioning as the predicate does; in other words, a noun (regardless of whether it functions as a subject, object, non-agreeing attribute, or adverbial modifier) has a stronger attraction with respect to an adverbial modifier than does a verb functioning as the predicate, if that adverbial modifier immediately follows the noun. 3.3. Infinitive attraction If, in a sentence, the predicate, expressed by a finite verb form, and the infinitive govern the same case, the infinitive attracts it more strongly than the predicate, expressed by a finite verb form. A finite verb form functioning as the predicate takes the governed case only when there is no infinitive, as in sentence (22); however, if there is an infinitive, see sentence (21), it attracts the governed case regardless of its position relative to the infinitive—before or after it. This property is especially clear when the governed case stands between the predicate (the finite verb form) and the infinitive; see sentence (23). (21) Then the entrepreneurs importing them should submit product quality certificates. (22) Then the entrepreneurs importing them should product quality certificates. (23) Then the entrepreneurs importing them should product quality submit certificates. 146 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka Thus, it can be argued that the syntactic attraction of the infinitive is stronger than that of the finite verb form functioning as the predicate with respect to the governed case. Sentences (24), (25), and (26) show that the governed case is attracted more strongly by the infinitive even when it stands before the predicate (finite verb form). Only when the infinitive is absent from the sentence does the predicate (finite verb form) take the governed case; see sentence (27). (24) To find him will be the task of Minister Eugenijus Gentvilas. (25) To find him will be the task of Minister Eugenijus Gentvilas. (26) Him the minister Eugenijus Gentvilas will find. (27) The minister Eugenijus Gentvilas will have him. 3.4. Definition of syntactic attraction between words Since the syntactic attraction between words is a recently observed phenomenon, no very detailed studies have yet been conducted, and it is not possible to provide a large amount of data about it. In the future, after syntactic annotation of the corpus has been completed, it may be possible to devise some kind of classification or even assign each word an appropriate syntactic attraction coefficient. For now, however, we can only describe the essence in general terms. First and foremost, the most important thing to say is that this property is relevant only when language is processed by computer, just as words are polysemous only in dictionaries—and the problem of their polysemy arises only when translating texts into other languages; in everyday spoken language, we never have any difficulty with word polysemy. (For several millennia, grandmothers in villages successfully spoke Lithuanian without even knowing a concept such as “word polysemy.”) When language is processed by computer, the sentence is viewed a little differently, and approached from a slightly different angle. Therefore, this work requires other tools as well. The essence could be stated as follows: syntactic attraction is a measure of words’ ability to realize their government. It differs greatly from valency. Everything works well, and valency is sufficient as long as there is one governing word in the sentence that can take a particular word (let us say, “the word under consideration”). But if there are two words in the sentence with the same government, both able to take the word under consideration, which one will take it after all, and which will remain without an expansion? Valency provides a comparison of governed words: some words are governed more strongly, others more weakly (i.e., it is necessary to take some words, but not others) (Sližienė 1994: 5–6); syntactic attraction is a characteristic of governing words. It compares two governing words and indicates a governing word’s power (or strength) to take a governed form, in comparison with another word in the sentence that has the same government (governs the same case). Valency does not go beyond the limits of a single verb (or another word that governs Straipsniai / Articles 147 DAIVA ŠVEIKAUSKIENĖ a form, such as an adjective); other words with government do not exist for it in the sentence. It takes no account of them, nor does it compare them with the word it describes, whose combinability it specifies. Syntactic attraction encompasses all the governing words in a sentence and compares them with one another in terms of their ability to realize their government. Definition: Syntactic attraction between words is the strength of their ability to take a governed case or an expansion. Syntactic attraction is a measure of governing words, a criterion for comparing them with one another. 4. SYNTACTIC ATTRACTION BETWEEN WORDS IN OTHER LANGUAGES The property of syntactic attraction between words should be more relevant to languages with free word order in the sentence. For comparison, corresponding sentences in three other languages—Russian, German, and English—are also presented. These languages differ greatly in the freedom of their word order, so it will be possible to see how the arrangement of certain parts of speech in a sentence affects syntactic attraction. 4.1. Russian To illustrate the claim in Section 3.1.1 that the syntactic attraction of an attribute expressed by a participle is stronger than the attraction of a predicate expressed by a participle with respect to an adverbial, the sentence Until then, for many people, especially the intelligentsia, only the “Great Soviet Encyclopedia” (“Bolshaya sovetskaya entsiklopediya”), published in Moscow, was available is presented in translation. (28) Until then, for many people, only the “Great Soviet Encyclopedia,” published in Moscow, was available. (29) Until then, for many people, only the “Great Soviet Encyclopedia” in Moscow was available. In this case, the Russian-language data coincide with the Lithuanian-language data. The claim in Section 3.1.3 that the attraction of a passive participle is stronger than the attraction of a noun with respect to the genitive can no longer be confirmed in Russian, because here the passive form governs a different case: the Lithuanian genitive is translated by the instrumental. The sentences This is stated signed by the minister in the report and This is stated in the minister’s report are translated by sentences (30) and (31). (30) This is stated in the minister-signed report. (31) This is stated in the minister’s report. 148 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka The claim in Section 3.2 that a noun attracts an adverbial immediately following it more strongly than the predicate can also be confirmed in Russian. This is evident both from the corresponding translations of the sentences He described in detail the exhibition in Kaunas, and in Kaunas the exhibition he described—sentences (32) and (33), in which the adverbial is expressed by the locative—as well as from the translations of the sentences Losses due to the disruption of exports through the Klaipėda terminal reached 28 million litas and Due to the disruption of exports through the Klaipėda terminal, losses reached 28 million litas—sentences (34) and (35), in which the adverbial is expressed by a prepositional construction. (32) He described in detail the exhibition in Moscow. (33) In Moscow he described the exhibition in detail. (34) Losses due to the interrupted export through the terminal reached 28 million. (35) Due to the interrupted export through the terminal, the losses reached 28 million. The claim in section 3.3 that the infinitive’s attraction is stronger than that of the predicate (the inflected verb form) with respect to the case it governs is also borne out in Russian. This time, the examples are the sentences Then the businesspeople importing them should submit product quality certificates and Then the businesspeople importing them should product quality certificates, and Then the businesspeople importing them should submit product quality certificates. Sentences in Russian with an analogous structure, illustrating the aforementioned property of syntactic relations, are used as examples; see sentences (36), (37), and (38). (36) He must return the book to me. (37) He must give me the book. (38) He must give me the book back. Thus, this property is also characteristic of Russian. Being close to Lithuanian in its freedom of word order, Russian has many similar syntactic properties; however, it cannot be claimed that the languages correspond completely. 4.2. German German word order is much stricter than that of Lithuanian or Russian, so a translation that follows the arrangement of words in a Lithuanian sentence becomes impossible. The strict requirement that the predicate occupy the second position makes it impossible to place a temporal adjunct and an object at the beginning of the sentence, before the subject; therefore, the sentences Until then, only the Great Soviet Encyclopedia, published in Moscow, was available to many, primarily to the intelligentsia, and Until then, the Great Soviet Encyclopedia, available only in Moscow, was accessible to many, primarily to the intelligentsia have no equivalents. The claim in Section 3.1.3 that the attraction of a passive participle is stronger than that of a noun with respect to the genitive cannot be confirmed for German, just as it cannot for Russian, since another case is used here too—in this instance, the prepositional construction von + Dativ. Straipsniai / Articles 149 DAIVA ŠVEIKAUSKIENĖ Another claim that cannot be tested because of the stricter word order in German is that a noun (whether functioning as subject or object) attracts an immediately following adjunct more strongly than a verb functioning as predicate (see Section 3.2). Nothing at all can be said about the attraction of an adjunct expressed by a prepositional construction in German, since it is impossible to provide a sentence Due to the disruption of exports through the Klaipėda terminal, losses reached 28 million litas in German while retaining an analogous word order. The claim in Section 3.3 that the attraction of an infinitive is stronger than that of the predicate (a finite verb form) with respect to a governed case is borne out in German; that is, the predicate takes a governed case only if there is no infinitive in the sentence. Analogues of the sentences Then the entrepreneurs importing them should have product quality certificates and Then the entrepreneurs importing them should submit product quality certificates are given by considering only structural correspondence (I have this book and I have to read this book), which illustrates the syntactic property under consideration; see sentences (39) and (40). (39) Ich habe dieses Buch zu lesen. (40) Ich habe dieses Buch. Because of the specific features of German grammar, it is impossible to provide a sentence with an analogous structure, for example, one equivalent to the sentence Then the entrepreneurs importing them should submit product quality certificates (I have to read this book). This is prevented by the characteristic German sentence bracket, which requires the predicate’s finite element and lexical verb to enclose all the words following the second (predicate) position. 4.3. English In English, it is relatively difficult to compare the attraction of a participle functioning as an attribute and that of a predicate expressed by a participle with respect to an adverbial modifier, because it is impossible to provide a translation corresponding to the word order of the Lithuanian sentence Iki tol daugeliui, pirmiausia inteligentijai, buvo prieinama tiktai Maskvoje leidžiama „Didžioji tarybinė enciklopedija“. In English, the subject cannot follow the predicate. As regards the differences in attraction between a passive participle and a noun with respect to the genitive (see Section 3.1.3), the same can be said of English as of Russian and German: a different case is used here, in this instance the prepositional construction by + noun. We can also see in English that a noun attracts a immediately following adverbial modifier more strongly than a predicate does (see Section 3.2). Translations of the sentences Jis išsamiai aprašė parodą Kaune and Kaune jis išsamiai aprašė parodą are given in examples (41) and (42). The same applies to an adverbial modifier expressed by a prepositional construction—the translations of the sentences Nuos- 150 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka toli dėl sutrikusio eksporto per Klaipėdos terminalą pasiekė 20 milijonų litų ir Dėl sutrikusio eksporto per Klaipėdos terminalą nuostoliai pasiekė 20 milijonų litų vertimai pateikti (43) ir (44) pavyzdžiuose. (41) He described the exhibition in London in detail. (42) In London he described the exhibition in detail. (43) Damage because of discontinuing of export through the terminal amounted 20 million dollars. (44) Damage amounted 20 million dollars because of discontinuing of export through the terminal. With regard to the syntactic attraction of the infinitive, as in German, two-sentence equivalents can also be provided in English. Sentences (45) and (46) do not contradict the claim in Section 3.3 that the predicate attaches a noun when there is no infinitive. (45) He has to read this book. (46) He has this book. 5. CONCLUSIONS To summarize the review of the sentences, we can draw conclusions that are more tendencies than strict claims, since only a few observations, rather than comprehensive studies, have been described. When Lithuanian is computerized, problems arise that are irrelevant in human communication. Therefore, further theoretical research on Lithuanian is needed. Observing errors made by the computer during automatic syntactic analysis led to the hypothesis that syntactic attraction between words exists, manifesting itself especially in cases where a sentence contains two words with the same government, but only one governed case. How can one determine which governing word will form a direct syntactic relation with that case, and which will remain without an expansion? Syntactic attraction between words differs fundamentally from valency: valency examines the combinability of one word and does not address the question of another word with the same government. In the examples examined, the syntactic attraction of a participle functioning as an attribute is most prominent. Syntactic attraction was also observed for nouns (regardless of whether they function as subjects or objects) and for infinitives. A comparison of analogous cases with other languages cannot be considered comprehensive, since equivalent translations were not always possible because of free word order. All the problems described in this article can also be viewed differently. From a linguistic perspective, one can say that in the sentence Aš turiu knygą, the word turiu is an independent predicate, whereas in the sentence Aš turiu perskaityti knygą, it is only a copula. This is clear to people, but a computer cannot know that in one sentence the same word is Straipsniai / Articles 151 DAIVA ŠVEIKAUSKIENĖ different in one instance and different in another, even though their spelling does not differ at all. This is where one should begin looking for formal features that would enable a computer to determine the type of predicate unambiguously from contextual data. Here, as in other cases, the syntactic attraction property of words could be helpful. A generalization from the examples presented reveals a tendency for the concept of syntactic attraction between words to be more important and more necessary in languages with free word order. In English, a word’s syntactic function is determined by its position in the sentence, and the property of word attraction can apparently do little to help identify sentence constituents. This article invites all Lithuanian linguists to join the discussion. Perhaps linguists will find it easier to notice other cases of word attraction that were not mentioned in this article. It discusses only what was observed in computer processing of language—only hypotheses which, of course, may prove unfounded. LITERATŪRA Allen James 1987: Natural Language Understanding. Amsterdam: The Benjamin/Cummings Publishing Company. Ambrazas Vytautas 1997: Dabartinės lietuvių kalbos gramatika. Vilnius: Mokslo ir enciklopedijų leidybos institutas. Daudaravičius Vidas 2012: Teksto skaidymas pastoviųjų junginių segmentais. Daktaro disertacijos santrauka. Kaunas: Vytauto Didžiojo universitetas. Daudaravičius Vidas 2002: Lietuvių kalbos sakinio sintaksinė analizė. Informatikos magistro darbas. Kaunas: Vytauto Didžiojo universitetas. Grigonytė Gintarė 2006: Priklausomybių gramatikos taikymas lietuvių kalbos apdorojime. Magistro darbas. Kaunas: Kauno Technologijos universitetas. Hellwig Peter 2003: http://www.cl.uni-heidelberg.de/~hellwig/dug-2003.pdf Ivič Milka 1973: Lingvistikos kryptys. Vilnius: Mintis. Jensen Karen, Heidorn George Emil, Richardson Stephen D. 1993: Natural language processing: The PLNLP Approach. Boston/London: Kluwer Academic Publishers. Mondal Prakash 2010: Exploring the N-th Dimension of language. – Natural Language Processing and its Applications. Research in Computing Science 46, ed. by Alexander Gelbukh. Mexico: Instituto Politecnico Nacional, 55–66. Nirenburg Sergei 1987: Machine Translation: Theoretical and Methodological Issues. London: Cambridge university Press. Sirtautas Vytautas 1978: Pagrindinių sakinio dalių derinimas. Vilnius: Lietuvos TSR aukštojo ir specialiojo vidurinio mokslo ministerijos Leid ybinė redakcinė taryba. Sližienė Nijolė 1994: Lietuvių kalbos veiksmažodžių junglumo žodynas 1. Vilnius: Mokslo ir enciklopedijų leidykla. 152 Acta Linguistica Lithuanica LXIX Sintaksinė žodžių trauka S l i ž i e n ė N i j o l ė 2004: Lietuvių kalbos veiksmažodžių junglumo žodynas 2 (2). Vilnius: Lietuvių kalbos instituto leidykla. Š v e i k a u s k i e n ė D a i v a 2005: Graph Representation of the Syntactic Structure of the Lithuanian Sentence. – Informatica 16(3), 407–418. W i n o g r a d T e r r y 1983: Language as a Cognitive Process. 1. Syntax. London: Addison-Wesley Publishing Company. Б е л е ц к а я И. П. 1983: Деревья зависимостей как инструмент синтаксического анализа текста. – Международный семинар по машинному переводу. Москва: ВЦП. Б и л а н В. Н., К р и с е в и ч В. С. и др. 1976: Семантико-синтаксический перевод в группе Статистика речи. – Автоматический анализ текста. Минск: МГПИИЯ. I N T E R N E T O N U O R O D O S 1. http://www.lki.lt/LKI_LT/index.php?option=com_content&view=article&id=803&Itemid=153 2. http://donelaitis.vdu.lt/main.php?id=4&nr=1_1 Syntactic Pull of Words S U M M A R Y By creating the automatic syntactic analysis of Lithuanian language a new phenomenon of language was perceived. If two words of the sentence require the same case as extension, which of them will get the syntactic relation with that case ant which remains without extension? If the flexible form of a verb, which performs the syntactic function of predicate controls accusative and participle, which performs the syntactic function of attribute, controls accusative too and that case is located in the sentence between the verb and participle, accusative is syntactic related with the participle. This power to make the syntactic relation with the required case was named syntactic pull of a word. The pull of participle, infinitive and noun are described and equivalent sentences in Russian, German and English are compared. The definition of syntactic pull of words is given. Įteikta 2013 m. balandžio 17 d. DAIVA ŠVEIKAUSKIENĖ Lietuvių kalbos institutas Petro Vileišio g. 5-216, LT-10308 Vilnius, Lietuva [email protected] S t r a i p s n i a i / A r t i c l e s 153