- Studies in Anaphora ebook | tisiwete.cf
- Anaphoric Annotation
- Anaphora and Distributivity: A study of same, different, reciprocals and others
Anaphora and Semantic Interpretation. By Tanya Reinhart. Edition 1st Edition. First Published Imprint Routledge. Responses toWh-Questions in English Conversation. Do we still need an army like in the First World War? An argumentative analysis of a television debate on abolishing compulsory military service in Switzerland. Laury, Ritva Levinson, Stephen C. Li, Wendan Topic Chains in Chinese Discourse. Anaphoric choices in Japanese fictional novels: The discourse arrangement of noun phrases, zero and third person pronouns.
Text - Interdisciplinary Journal for the Study of Discourse Cognitive Linguistics Rosen, Nicole Demonstrative Position in Michif. Grammatical roles, Coherence Relations, and the interpretation of pronouns in Chinese. Segment 1, below, chosen for its simplicity and lack of ambiguity, is used to illustrate the basic concepts underlying this process.
Segment Houses i are bought because they i offer comfort. The pronoun they does not relate directly to an object in the world but to a word that was mentioned before. The mental task performed by the reader, when processing this sentence, is to go back in the text and find the word it relates to. In Segment 1, there are four words preceding the pronoun because, bought, are and houses , but only one is a serious candidate houses. The pronoun they can only be replaced by a plural noun and the only word that fulfills this condition is houses.
Examples in real-life situations are not always so straightforward. One complication that may arise is the possibility of more than one rightful candidate for the antecedent, as demonstrated in the following case:. Segment Houses i are bought by people because they offer i comfort. Now there is not one but two candidates for they , which are the words houses and people both plural nouns.
How to solve this problem? One hypothesis is that it can be solved by applying syntactical constraints. It can be argued that there is a syntactic parallelism between the noun houses and the pronoun they , that is, both houses and they are in the subject position in their own clauses. The word people , on the other hand, although a plural noun, does not share this parallelism with they. Thus, between the two candidates, we choose the noun houses. Syntactic constraints based on parallelism, however, seem to work fine only as far as the examples are carefully chosen.
In Segment 2, for example, simply changing one lexical item for another may totally revert the relationship between the anaphor and the referent. This can be seen in the following Segment, where offer is replaced by like. Segment Houses i are bought by people j because they j like comfort. Again there are two candidates, which are exactly the same as in the previous Segments. Syntactic parallelism, which so efficiently facilitated the choice between the two candidates in the previous Segment, does not seem to work any longer.
Studies in Anaphora ebook | tisiwete.cf
The only candidate that can rightfully be in the position occupied by they is people. The other candidate houses violates a semantic constraint: houses do not like things; only people like things. Syntactic parallelism can, thus, be overruled by semantic constraints; it is not enough for the antecedent to possess the same syntactic function as the anaphor; both antecedent and anaphor must also share semantic features.
Syntactic and semantic constraints, still, are not enough in solving the problems associated with anaphora resolution, as can be seen in the following cases. Segment The companies i sold their cars j to the sheiks k because they i offered long-term guarantee. Segment The companies i sold their cars j to the sheiks k because they j were bulletproof models. Segment The companies i sold their cars j to the sheiks k because they k offered more money. Segments , apparently, can only be solved by resorting to a representation of the world in which there are sellers, buyers, and commodities that change hands: money from buyers to sellers and cars from sellers to buyers.
We also need to know that cars can be bulletproof, that companies usually offer guarantees on what they sell and that sheiks can be very rich.
- Chechens: A Handbook (Causasus World, Peoples of the Caucasus).
- Breakup of Liquid Sheets and Jets.
- Military Leadership in the British Civil Wars, 1642-1651;
- Abandoned: The Untold Story of the Abortion Wars!
All this world knowledge has to be readily available for the antecedent of they to be correctly identified in each of the Segments. The problem, however, in using world knowledge is its computational cost. There are so many variables that a combinatory explosion is inevitable. Each variable can interact with so many other variables, with so many possibilities of different combinations that the system may enter an endless loop and the right combination is never encountered.
- Natural Polymers vol.1.. Composites.
- Anaphora Resolution.
- Dynamics of Political Violence: A Process-Oriented Perspective on Radicalization and the Escalation of Political Conflict!
The solution to the problem of tracking down the antecedent in anaphora seems to lie somewhere between the simplicity of syntactic constraints and the complexity of world knowledge. This is the problem addressed in this investigation. There are two questions to be answered here: 1 what are the limitations of syntactic constraints in anaphora resolution?
Discourse, cognition and textual constraints. Anaphora can be studied from different perspectives, including discourse e. McEnery and Botley, ; Indursky, , cognition e. Langacker, ; van Hoek, and textual constraints Dagani and Itai, ; Nasukawa, ; Mitkov and Belguith, Many of these studies emphasize the correlation between certain discourse-pragmatic factors e. Fox summarizes these correlations as follows:. Fox, vii. Reference-tracking devices such as the use of pronouns, zero anaphors or full nouns, even when correlated to topicality, discourse sequences and speaker's cognitive or affective states, do not reveal very much in terms of the process involved.
All it amounts to, in fact, is the probability of which anaphor to use - on a scale that ranges from zero to a full noun. There is no description of what really happens in the mind when the reader or listener finds an anaphor and tries to track down its antecedent, inside or outside the text. Anaphora resolution at this low level of processing, most of it below conscious control, is probably not an area of interest to the discourse-pragmatic paradigms of research, which may be more concerned with the general picture, viewing the process from a more abstract level.
A very different perspective is offered by studies in computational linguistics, where the implementation of an anaphora-resolution system makes it necessary to translate abstract concepts into machine-readable code, using data that have to be found on the textual surface. With the processing power of modern computers, the variety of these data have been increased. We are no longer limited to low-level types of linguistic data, such as part of speech information, but we can also include higher-level linguistic structures, related to possible configurations of relationships between different segments of the text.
We can recursively encapsulate chunks of language into ever-increasing units, building larger blocks, and abstracting their characteristics. The crucial point, however, is that the link between the anaphor and the antecedent should be unambiguous, with total agreement between different readers consuming the same text. Should disagreement arise, not due to differences in the texts but to differences in readers' interpretations, the problem is beyond solution from the perspective of computational linguistics, which is basically algorithmic. Attempts to endow computers with the world knowledge needed to attribute meanings to text, instead of only extracting meaning from it, are theoretically interesting but extremely costly and, for the time being, unfeasible for practical purposes.
Anaphora resolution, in terms of computational linguistics, cannot be attributed to the cognitive or affective states of the readers; the data have to be present on the surface of the text. Linguistic data that can be found on the text such as gender and number agreement, c-command restrictions, syntactic parallelism, lexical repetition, or antecedent proximity are favored in the resolution process because they can be more easily dealt with by the available tools in computational linguistics. These tools usually rely on the concepts of "constraints" and "preferences" - where "constraint" is the more powerful of the two devices.
Knowledge-poor solutions, using corpus-driven methodologies, statistical and probabilistic models, are substantially preferred. Some strategical approaches for tracking the antecedent, as opposed to purely statistical models, have also been proposed. These approaches can be formalized in terms of rules, usually based on constraints and preferences. The following preferences, for example, can be used to select the antecedent based on Mitkov , :. Paraboni also adopts a strategical approach, using a combination of constraints and preferences in his study of Portuguese anaphora involving possessive adjectives.
Anaphora and Distributivity: A study of same, different, reciprocals and others
These adjectives, when belonging to the third person, are interesting in Portuguese because as anaphors they do not agree in number and gender with the antecedent - as they do in English, for example - but with the thing being possessed, a feature that makes it more difficult to locate the antecedent. Reference-tracking strategies, therefore, have fewer constraints and preferences to rely on. One of the most productive was the presence of a coordinative conjunction between the anaphor and the NP, as shown in Segment 7.
Segment 7: The law i and its i consequences. Paraboni, himself, however, is cautious to point out, that exceptions to this rule can be easily found. This seems to be the case, for example, with complex NPs such as 8 and 9 , where the coordinative conjunction rule is overridden by semantic constraints See also Baltazart and Kister, Segment 8: The book i on divorce j and its j consequences. Segment 9: The book i on divorce j and its i author. In summary, it seems that a description of what happens in anaphora resolution falls into two extremes, offering two alternatives.
On one hand, the process is analyzed in very general terms, from a highly abstract level, providing the whole picture, but failing to offer a description of crucial aspects of the process, which are implicitly acknowledged, but not explicitly detailed. On the other extreme, the process can be minutely described in one of its aspects, analyzing every little step, but then failing to provide the whole picture. This investigation opts for a third alternative, relying heavily on the data available from the surface of the text, but, looking at them both horizontally from a syntagmatic perspective and vertically in a more paradigmatic way , where semantic relations between the lexical items that make up the text are also taken into account.
In Segments 8 and 9 above, for example, it can be argued that the lexical item author shares more semantic features with book than divorce - which should offer a cue for solving the problem of finding the right antecedent for the possessive its. The main purpose here is to test a procedure that can be summarized in two steps: the first is to select an antecedent that complies with syntactic constraints including gender, number, syntactic parallelism, etc. Any project in the area of computational linguistics, including information retrieval, dialogue processing, and machine translation, has to allocate a major part of the system to solve this problem.
The decision at what stage of the process to attack the problem is dependent on many aspects, including the theoretical approach being used. For the approach proposed here, based on a machine translation project between English and Portuguese, anaphora is dealt with after some preliminary analysis has already been performed on the text being processed, including the following:. Part of speech assignment: each word in the text has already been classified into one of the basic word classes noun, verb, adjective, etc. Noun phrase segmentation: complex nouns such as combinations of two nouns stone house , adjective and nouns the big house have already been segmented with identification of the corresponding headword.
Case assignment: the syntactic function nominative, accusative, dative, etc. Table 1 shows how two NPs are classified. Notice that a large house on the hills is classified as singular NP, since all the words that make up the noun phrase are governed by the headword house. For this investigation the pronoun "they" was chosen. There is a theoretical and a practical reason for this choice. In terms of theory, it is expected that the analysis will help explain the interrelation between anaphora and text, from a strictly linguistic point of view.
The question asked here is whether or not it is possible to resolve anaphora without resorting to world knowledge - basically how far anaphora is dependent on syntactic and semantic constraints. In terms of practice, the results could be immediately applied to machine translation from the English language into many romance languages such as French, Spanish or Portuguese, where the pronoun they has different translations, depending on the antecedent. The basic methodology involved a selection of 1, occurrences of they from a corpus of 10 million words of explanatory texts.
For this selection a concordancer program was used. This type of program allows for the occurrence of a given word or combination of words to be automatically extracted from the corpus and listed according to a selected order alphabetically by first left word, first right word, second word, etc.