{ "paper_id": "1991", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T07:34:14.677071Z" }, "title": "PARSING = PARSIMONIOUS COVERING?", "authors": [ { "first": "Venu", "middle": [], "last": "Dasigi", "suffix": "", "affiliation": { "laboratory": "", "institution": "Wright State University Research Center", "location": { "addrLine": "3171 Research Boulevard Dayton", "postCode": "45420", "region": "OH" } }, "email": "vdasigi@cs.wright.edu" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "Many researchers believe that certain aspects of natural language processing, such as word sense disambiguation and plan recognition in stories, constitute abductive inferences. We have been working with a specific model of abduction, called parsi monious covering, applied in diagnostic problem solving, word sense disambiguation and logical fonn generation in some res tricted settings. Diagnostic parsimonious covering has been extended into a dual route model to account for syntactic and semantic aspects of natural language. The two routes of covering are integrated by defining \"open class\" linguistic concepts, aiding each other. Toe diagnostic model has dealt with sets, while the extended version, where syntactic con siderations dictate word order, deals with sequences of linguistic concepts. Here we briefly describe the original model and the extended version, and briefly characterize the notions of covering and different cri teria of parsimony.-Finally we examine the question of whether parsimonious covering can serve as a general framework for pars ing.", "pdf_parse": { "paper_id": "1991", "_pdf_hash": "", "abstract": [ { "text": "Many researchers believe that certain aspects of natural language processing, such as word sense disambiguation and plan recognition in stories, constitute abductive inferences. We have been working with a specific model of abduction, called parsi monious covering, applied in diagnostic problem solving, word sense disambiguation and logical fonn generation in some res tricted settings. Diagnostic parsimonious covering has been extended into a dual route model to account for syntactic and semantic aspects of natural language. The two routes of covering are integrated by defining \"open class\" linguistic concepts, aiding each other. Toe diagnostic model has dealt with sets, while the extended version, where syntactic con siderations dictate word order, deals with sequences of linguistic concepts. Here we briefly describe the original model and the extended version, and briefly characterize the notions of covering and different cri teria of parsimony.-Finally we examine the question of whether parsimonious covering can serve as a general framework for pars ing.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "Natural languages are rife with ambi guity. There are lexical ambiguities; words in isolation may be seen to have multiple syntactic and semantic senses. There are syntactic ambiguities; the same sequence of words may be viewed as constituting different structures. And finally, there are semantic and pragmatic ambiguities, all of which may be resolved in context. Ambi guity and its context-sensitive disambigua tion, it turns out, are two important charac-teristics of abductive inferences.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "There have been various attempts at \u2022 characterizing abductive inference and its explanatory nature [Appelt, 90; Charniak and McDermott, 85; Hobbs, et al., 88; Josephson, 90; Konolige, 90; Pople, 73; Reggia, 85; etc.] . While they diffe r some what in details, they all boil down to accounting for some obseIVed features using potential explanations consistently in a \"parsimonious \" (often \"minimal\") way. Over the past decade, a formal model for abduction based on these ideas was developed at Maryland; this theory is called parsimonious covering. Toe theory ori ginated in the context of simple diagnostic problems, but extended later for complex knowledge structures involving chaining of causal associations.", "cite_spans": [ { "start": 100, "end": 108, "text": "[Appelt,", "ref_id": null }, { "start": 109, "end": 112, "text": "90;", "ref_id": null }, { "start": 113, "end": 136, "text": "Charniak and McDermott,", "ref_id": null }, { "start": 137, "end": 140, "text": "85;", "ref_id": null }, { "start": 141, "end": 147, "text": "Hobbs,", "ref_id": null }, { "start": 148, "end": 155, "text": "et al.,", "ref_id": null }, { "start": 156, "end": 159, "text": "88;", "ref_id": null }, { "start": 160, "end": 170, "text": "Josephson,", "ref_id": null }, { "start": 171, "end": 174, "text": "90;", "ref_id": null }, { "start": 175, "end": 184, "text": "Konolige,", "ref_id": null }, { "start": 185, "end": 188, "text": "90;", "ref_id": null }, { "start": 189, "end": 195, "text": "Pople,", "ref_id": null }, { "start": 196, "end": 199, "text": "73;", "ref_id": null }, { "start": 200, "end": 207, "text": "Reggia,", "ref_id": null }, { "start": 208, "end": 211, "text": "85;", "ref_id": null }, { "start": 212, "end": 217, "text": "etc.]", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "A diagnostic problem specified in tenns of a set of obseIVed manifestations is solved in parsimonious covering by satisfying the coverage goal and the goal of parsimony. Satisfying the coverage goal requires accounting for each of the obseIVed manifestations through the known causal associations. Ambiguity arises here, because the same manifestation may be caused by any one of several candidate disorders. Ensuring that a cover contains a ' 'parsimonious'' set of disorders satisfies the goal of parsimony. There could poten tially be a large number of covers for the observed manifestations, but the ''parsi\ufffd monious'' ones from among them are expected to lead to more plausible diag noses. The plausible account for a manifes tation may be one disorder in one context and another disorder in a different context. Such contextual effects are to be handled automatically by the specifi c criterion of parsimony that is chosen.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "For medical diagnosis, reasonable cri teria of parsimony are minimal cardinality, irredundancy and relevancy [Peng, 85] . Minimal cardinality says that the diagnosis should contain the smallest possible number of disorders that can cover the observed symptoms. A cover is considered irredun dant (not redundant) if none of its proper subsets is also a cover, i.e., if the cover contains no disorder by removing which it can still cover the observed symptoms. Relevancy simply says that each disorder in the cover should be capable of causing at least one of the observed manifestations.", "cite_spans": [ { "start": 109, "end": 115, "text": "[Peng,", "ref_id": null }, { "start": 116, "end": 119, "text": "85]", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "Consider an abstract example where disorder d 1 can cause any of the manifestations m 1 and m 2 ; d 2 can cause any of m 1 , m 2 and m 3 ; d 3 can cause m 3 ; d 4 can cause m 3 and m 4 ; and finally, d 5 can cause m 4 \u2022 If the manifestations { m 1 , m 2 , \ufffd} were observed, the disorder set { \ufffd } constitutes a minimal cardinality cover; the irredundant covers that are not minimal cardinality cov ers are { d 1 , d 3 } and { d 1 , d 4 }; and an example of a redundant, _but relevant cover would be", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "{ d 1 , d 3 , d 4 }. While { d 2 , d 5 }", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "is a cover that has an irrelevant disorder (d 5 ) in it, { \ufffd ' d 4 } is a non-cover, since together the disorders in this set cannot account for all observed manifestations.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "Several natural language researchers have been actively involved in modeling abductive inferences that occur at higher levels in natural language, e.g., at the prag matics level. Abductive unifications that are required in perfonning motivation analysis, for instance, mightcall for making the least number of assumptions that might potentially prove false [Chamiak, 88] . Lit man uses a similar notion of unification, called consistency unification [Litman, 85] . Hobbs and his associates propose a method that involves minimizing the cost of abduc tive inference where the cost might involve several different components [Hobbs, et al ., 88] . Although [Charniak and McDermott, 85] indicate that word sense disambiguation might be viewed as abductive, nobody has pursued this line of research. It is very clear that there exists a strong analogybetween diagnostic parsimonious covering and concepts in natural language process ing.", "cite_spans": [ { "start": 357, "end": 366, "text": "[Chamiak,", "ref_id": null }, { "start": 367, "end": 370, "text": "88]", "ref_id": null }, { "start": 450, "end": 458, "text": "[Litman,", "ref_id": null }, { "start": 459, "end": 462, "text": "85]", "ref_id": null }, { "start": 623, "end": 630, "text": "[Hobbs,", "ref_id": null }, { "start": 631, "end": 639, "text": "et al .,", "ref_id": null }, { "start": 640, "end": 643, "text": "88]", "ref_id": null }, { "start": 655, "end": 679, "text": "[Charniak and McDermott,", "ref_id": null }, { "start": 680, "end": 683, "text": "85]", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "There are, however, important differences as well. These similarities and differences are summarized in Table I . We have tried to extend parsimonious cov ering to address some of the idiosyncrasies of language ( contrasted to diagnosis) and apply it to low level natural language pro cessing.", "cite_spans": [], "ref_spans": [ { "start": 104, "end": 111, "text": "Table I", "ref_id": null } ], "eq_spans": [], "section": "Introduction", "sec_num": "1." }, { "text": "Linguistic concepts are viewed in par simonious covering to be much like disord ers and manifestations in diagnostic prob lems. However, in order to account for word order and structural constraints in language on the one hand and to account for the lexical and semantic content on the other, two aspects are attributed to each linguistic concept These two aspects are loosely referred to as syntactic and seman tic aspects, respectively. Concepts are covered parsimoniously in these two aspects, and the processes of covering are called syntactic and semantic covering. The categori\ufffds sh9wn jn bold face are mandatory categories, i.e., categories that must be present for' the description to viably apply to a context.. Semantic con-. siderations govern whether a \u2022 catego_ry is mandatory in a description:\u2022 Depending on the domain, ''the patient blind'' might still make sense (indicating that the -omitted copula is . not mandatory), but ''the patient' \u2022 alone does not make complete sense (indi cating that for this type of sentences, an adjectival complement is\u2022 -mandatory). See [Dasigi, 88] for discussion. Suppose the input sequence is . Some valid covers (covering sequences) are , , , , , etc. Some non covers are , , , ,", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Covering and Parsimony in Language", "sec_num": "2." }, { "text": "etc., either because they cannot account for all the categories in the input sequence or because they cannot account fo r the correct order. Note that although is a cover, is not a cover. For instance, it makes sense to cover '' paint the wall\" with the sequence , but not by . Irredundant covers include and . Of these two irredundant covers, the former is also minimal (i.e., of minimal cardinality) and the latter is not. Insertion of c 5 into any valid cover causes it to be a non-viable cover since the category man datory to c 5 , namely, w 7 is not present in the input sequence to be covered. Thus, is a non-viable cover.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Covering and Parsimony in Language", "sec_num": "2." }, { "text": "\u2022 Consider the cover .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Covering and Parsimony in Language", "sec_num": "2." }, { "text": "Superficially, it appears to be a redundant cover since c 1 by itself is a cover. When the second rather than the first description of c 1 is taken into account, however, there is no redundancy in the cover, in a certain sense. For more concreteness, consider the following two classic sentences that differ in a single word:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Covering and Parsimony in Language", "sec_num": "2." }, { "text": "\"John painted the wall with a crack.\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Covering and Parsimony in Language", "sec_num": "2." }, { "text": "'' John painted the wall with a brush. '' Now, suppose there exist the usual descrip tions . for noun phrases (Noun-Phrase) and prepositional phrases (Prep-Phrase). Although in both sentences, the highlighted words can be syntactically covered by the irredundant cover , the sequence is a more appropriate cover in the second sen tence, and we would like to consider that cover as irredundant, too. This characteri zation of irredundancy is obviously impor tant, and is somewhat .similar to the notion of \"relevant diagnostic . covers\" defined in the previous section. ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Covering and Parsimony in Language", "sec_num": "2." }, { "text": "A significant prototype was imple ment\ufffd to apply this algorithm in the con text of an interface to an expert system. Instead of syntactic categories such as nouns, verbs, noun-phrases, etc., sem antic categories were used in the syntactic c\u2022 over ing process. Semantic covering was per fonned using domain-specific concepts defined in a knowledge base used by the expert system. In an OPS5-style expert system language, domain-specific concepts such as, patient, vision, blind, etc. were classified into semantic categories such as objects (obj), attributes (attr), values (val), etc. Two application domains were con sidered; the first domain is characterized by a sizable, prototype neurological knowledge base and the other deals with a toy chemi cal spills knowledge base. Some examples that were successfully handled by the pro totype interfaces are:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "' These examples demonstrate the use of lex ical information, limited ability to handle ungrammatical sentences, interpretation of sentences in a discourse context rather than in isolation, etc. Note that the first fe w words of the first two inputs are the same. Their interpretations are, however, significantly distinct in the . context of the knowledge base that was used, illustrating a form of non-monotonic inference in text interpretation. All but the last input is from the neurology domain and the last one is from the other.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "A very simple example of parsimoni ous covering is given below to convey the flavor of the approach. Details are omitted due to space considerations, and we appeal to the reader's intuition in making sense out of this brief example. Suffice it to say that the category assert (and its variations) corresponds to sentences or clauses; obj and attr (and their variations) correspond to noun phrases; and val (and its varia tions) correspond to noun phrases or adjec tive complements. The category asg-verb stands for \"assignment verb\" (e.g., \"is\"). There are different ways an assert may be described in terms of the other categories mentioned so far. Often, val is a mandatory category in describing an assert (that is, it is unlikely that an assert makes semantic sense if a val is not present). Now, suppose a sentence begins with \"Vision is ... \" and is to be covered syntactically. One sequence of terminal categories that cover the first two words in this sentence is (attr, asg-verb) among others, since vision is an attribute and the word \"is\" is an instance of asg\ufffdverb. Since, this is an embedded sequence of what is expected of the above description of assert, the category assert is postulated to be a non viable syntactic cover fo r the first two words. It is a cover because the two semantic categories occur in the description of assert, in . the correct order. But the cover is non-viable nevertheless, because, not all mandatory categories in this particu lar description, namely, val, have occurred yet When all expected , mandatory categories occur, the cover will be con sidered viable. Further, viable \u2022 or not, the cover is tentative because other possible covers exist and one of the other covers might prove to be globally more plausible. Now, suppose the sentence ends as fol lows:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": ". .. impaired.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "Then, since impaired is a domain-speci fie value, the mandatory category is also encountered; so \u2022 assert is confirmed as one of several viable syntactic covers for the given words. To keep things simple for the present purposes, it is assumed that assert turns out to be the most plausible syntactic cover.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "The covering category in this exam ple, namely assert, was designated as an open class category. In general, if the category that has just been postulated as a cover happens to be an open class category, it initiates semantic covering (with the standard notion of compositional ity), thus integrating the use of both (that is, syntactic and semantic) aspects of knowledge. Now, we continue the exam ple from the viewpoint of semantic co\ufffder ing. Recall, however, that this process is interleaved with syntactic covering, and does not necessarily follow it. See Figure 1 .", "cite_spans": [], "ref_spans": [ { "start": 560, "end": 569, "text": "Figure 1", "ref_id": "FIGREF2" } ], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "The word ''vision'' is covered, among other things indicated above, by a concept that has the semantic category attr. Category attr is of open class and so not surprisingly the concept that covers \"vision\" also has a domain-specific entity, say a12, that uniquely characterizes it. In effect, this one linguistic concept covering ''vision,'' has two facets: the semantic category attr and the domain-specific entity a12. Similarly, the word \"impaired\" is covered by, among others, a concept of the semantic category val that has the unique domain-specific entity, say v30, associated with itself. The verb \"is,\" however, is covered by a concept of the category asg verb and since asg-verb is a not an open class category, it does not have a corresponding domain-specific entity.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "As already explained in the course of syntactic covering, assert is computed to be a syntactic cover; it also turns out to be a parsimonious syntactic cover. For semantic covering, what needs to be covered is the set of entities grouped under this category, i.e., a12 and v30, by identi fying domain-specific associations that relate them. Definitions of parsimony and cover ing in . the semantic route attempt to cap ture these intuitions, and the concept characterized by the semantic category assert and the domain-specific entity con stituted by (attr=a12, val=v30) becomes the integrated parsimonious cover for the given sequence of words. A simple domain-specific entity may be represented by a single node in the semantic network, e.g., an attribute. Also, a non-atomic sub graph of the semantic network can represent a more complex domain-speci fie entity, e.g., an assertion that relates an attribute and a possible value for it. Either kind of domain-specific entity whether represented by a single node or _ by a subgraph in the domain-specifi c semantic network -is said to be covered by any of its supergraphs. Since any super-graph of a domain-specific concept can cover it, for any domain-specific concept there are potentially a huge number of covers, some of which are very redundant There should be some means of controlling the number and sizes of potential covers. Criteria of parsimony and other constraints are used to achieve this control. A criterion of parsimony called cohesiveness is chosen, inspired by the fact that in order to be understandable, text must be cohesively connected. A set of semantic categories are designated as asser tionals (loosely corresponding to the notion of a sentence or an independent clause in English). A semantic cover corresponding to a non-assertional category is considered to be cohesive if it is the smallest (in tenns of nodes) connected graph covering the concepts in question. A semantic cover corresponding to an assertional category is considered to be cohesive if either it is the smallest connected graph covering the concepts being covered or it is a not necessarily connected graph of several such domain-specific entities belong ing to assertional categories. If there is more than one unconnected cover for the same concepts, the smallest connected cover of such unconnected components is the cohesive cover. It can be seen that cohesiveness refers to the \"size\" of the covers, and it is similar to '' minimal cardi nality,\" used in early versions of parsi..: monious covering theory for diagnostic problems. Indeed, if minimality were to be extended to structured entities, it would be similar to cohesiveness above. Cohesive ness refers to how well a cover fits into its surrounding context, a generalization of the notion of minimal cardinality, applied to structured entities.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "Consider two consecutive concepts that have the same domain-specific entity (say an object) as one of the many candi date covers. Since both concepts can be covered by the same entity, the entity is a minimal cover for both of \ufffdem together. This example of parsimonious covering is essentially the same as minimal covering-in the unextended parsimonious covering theory for diagnostic problem solving. However, suppose the two concepts involved cannot be covered by the same domain-specific entity. A minimal cover in the unextended parsimonious _ covering theory would consist of any pair of entities (pair -because there are two words to be covered) such that each entity in the pair covers one concept. But when structured entities with semantic associations among them are considered, the entities in. the pair must also unify, taking domain-specific associations into account 3 \u2022 Unification of such structures corresponds to a searc11 in the domain-specific semantic network, say, by marker passing [Charniak, 83] .", "cite_spans": [ { "start": 1005, "end": 1015, "text": "[Charniak,", "ref_id": null }, { "start": 1016, "end": 1019, "text": "83]", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "One important remark about semantic covering is in order. Cohesiveness, as a notion of parsimony for semantic covering, is _ intended to capture how plausible a semantic cover is. But it is possible that a cohesive \ufffdover might -tum out to be implausible when checked for well formedness. Because of this possibility, there should be means to recompute the next most plausible (cohesive) cover. Thus, whenever a cohesive cover is found, all the irredundant covers must be saved so that the space of possibilities they consti tute can be explored for cohesiveness if the cohesive cover that was found \u2022 were to be rejected later. Consider the \u2022 following abstract example. Let x 1 , Xz, \ufffd' x 4 and x 5 be\u2022 the senses of one ambiguous linguistic concept and y 1 , y 2 , y 3 and y 4 be the senses of another concept. If these two concepts were syntactically covered together by an open class semantic category, _ then 3Tiris can be understood as follows: An assertion may be viewed as a predicate assert(?v,?a,?o) covered by a specific value v 1 and the other is covered by a specific attribute a2, the coven effectively specify the assertions \u2022assert(v t?aa,?oo) and usert(?vv,a2,?ooo) . respectively. Now unification may be performed in the usual sense. semantic covering will be initiated. Now, what needs to be semantically covered is the conjunction of the following two dis junctions (representing 5*4 = 20 combina tions): { x l Xz \ufffd-x ,. X5} and { y l Y2 Y3 Y4 } Suppose a cohesive cover is found between \ufffd and y 3 \u2022 Then the irredundant cover will be constituted by the following three conjunctions of disjunctions (which represent the remaining 19 combinations): { x l \ufffd x ,. X5} and { y l Y2 Y ,.} { x 2} and { Y1 Y2 Y4 } { x 1 \ufffd x 4 x 5 } and { y 3 } If the cohesive cover that was discovered gets rejected, the next most cohesive cover might be computed from these irredundant covers.", "cite_spans": [], "ref_spans": [ { "start": 993, "end": 1009, "text": "assert(?v,?a,?o)", "ref_id": null }, { "start": 1010, "end": 1184, "text": "covered by a specific value v 1 and the other is covered by a specific attribute a2, the coven effectively specify the assertions \u2022assert(v t?aa,?oo) and usert(?vv,a2,?ooo)", "ref_id": "FIGREF2" } ], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "The dual-route parsimonious covering algorithm uses a discrete marker passing scheme to find cohesive \u2022 semantic covers. One problem with irredundant syntactic covering is that typically there are too many such covers. (The advantage, how ever, is that all useful infonnation is always available.) Since there are too many candidate syntactic covers, there exists a need to focus search for the best ones.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "Consequently, the dual-route algorithm uses semantic criteria to select a candidate to be covered at the next layer. Thus, the algo rithm incorporates notions of parsimonious covering and best-first search to integrate syntactic and semantic processing towards the goal of synthesizing the final interpreta tion for an input text.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Some Examples", "sec_num": "3." }, { "text": "The ability of parsimonious covering to handle ungrammatical sentences, as exemplified earlier, does not call fo r any special ( or ad hoe) handling. It is a natural consequence of the very definition of cov ering itself. One could argue that a con ventional production rule approach may easily be augmented to achieve the same effect. For instance, it might be possible that a description such as: \ufffdert: attr asg-verb val,", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Discussion", "sec_num": "4." }, { "text": "where val is mandatory, can be encoded into the following production rules: assert --> attr asg-verb val I attr val", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Discussion", "sec_num": "4." }, { "text": "lv al I \u2022\u2022\u2022\u2022", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Discussion", "sec_num": "4." }, { "text": "the number of such rules can grow exponentially in the number of non mandatory categories. The previous paragraph should not be misconstrued as downplaying the significance of syntax in language. Indeed, the verb is plays a crucial role in disambi guating sentences such as, \"Flying planes is/are dangerous.\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Discussion", "sec_num": "4." }, { "text": "Our point is that omission of the copula in such sentences still does not make them incomprehensible. It does leave the sen tence ambiguous, to be sure. At present, the semantic covering process does not worry about number agreement between the verb and subject, unless ambiguity arises. The underlying assumption here is that people try to make sense, and are not always grammatical. 1 In summary, parsimonious covering provides a framework to view parsing natural language as an abductive process. A proof of concept is provided by implement ing the basic ideas in an application independent interface shell. Admittedly, the semantic knowledge used is very restricted in nature, at the moment appropriate onlt to an object-oriented class of applications. The presumed logical fonn is also, correspondingly, of a limited generality. Many significant linguistic issues remain to be answered in this framework, however. Two features of this preliminary work (namely, use of a semantic grammar-like descriptio1 that are closely related to the class of ex 1 pert systems for which interfaces could be generated, and reliance 'on the assumption that ambiguity resulting from ungrammaticality is resolvable in context) make it hard to predict the generality of the techAfque for unrestricted natural language. It is hoped that planned exten sions, in the directions of using regular syntactic categories, and incorporation of further structure into verb definitions ( con sequently lmaking the logical form much more general), might help answer these important questions.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Discussion", "sec_num": "4." }, { "text": "This notion is very similar to that of open class words in languages. Non\ufffd class concepts only have syntactic as pect, and correspond to \"syntactic sugar\" in language. Sec [Duigi, 88) for more discussion.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The majority of. test inputs used by the prototyp came from physician,\u2022 anonymous case descriptioos, where insuring the grammaticality of sentences was, apparently, not the fint", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [ { "text": "The ,author acknowledges the suppott received from the State of Ohio Research Challenge l grant that enabled him, in part. to prepare this paper. Past support from Jim Reggia of the University of Maryland is also gratefully acknowledged.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Acknowledgements", "sec_num": null } ], "bib_entries": {}, "ref_entries": { "FIGREF0": { "uris": null, "type_str": "figure", "num": null, "text": "Semantic covering also involves the notions of covering and parsimony, where parsimony considerations indicate the plausibility of semantic covers." }, "FIGREF1": { "uris": null, "type_str": "figure", "num": null, "text": "For the sake of completeness, we briefl y describe the salient fe atures of semantic covering. A detailed account and algorithms may be found in [Dasigi, 88). The conceptual objects manipulated by semantic covering are domain-specific semantic senses. For semantic covering, the order of the concepts being covered is no longer important Semantic covering involves discovering the relationships under lying the domain-specific entities evoked by input words, so that a parsimonious seman tic cover can be synthesized for them; this cover corresponds to the logical fonn of the original sequence of words. There are two types of semantic covering. The first type of covering involves covering indivi dual content words by domain-specific senses corresponding to objects, attributes, etc. This type of covering involves only lexical associations. Here, a domain-specifi c entity semantically covers a content word if any of the content words in the name or synonyms of the entity is morphologically related to the word itself or a domain specific or domain-independent synonym of the word. The other type of semantic covering is based on the relationships in a domain specific semantic network." }, "FIGREF2": { "uris": null, "type_str": "figure", "num": null, "text": "Interleaving of syntactic and semantic covering. 1be dashed arrows indicate other concepts that are evoked, e.g., other attributes named by \"vision,\" other types of verbs that \"is\" evokes and many other concepts named by ''impaired.''" }, "TABREF0": { "html": null, "type_str": "table", "content": "
Parsimonious Covering Theory (Dia osis)Natural Language Processing
SIMILARITIES:
symptoms disorders intennediate syndromes symptoms with multiple causes \u2022pathognomonic Irianif estations\u2022 observed manifestations (to be explained) causal relation (between symptoms and disorders) diagnostic explanation (i.e.\ufffd a set of disorders)words internal \ufffdrtions word senses and structures ambiguous words unambiguo_ us words input text (sequences of words) (to be interpreted) lexical and semantic associations (between words and senses) -\u2022 \u2022 , \u2022 . < semantic interpretation_ ,-, ; '._ : :\u2022 ;. (i.e., a set of related assertions)
DIFFERENCES:
order of entities ignored sets of entities only \u2022one , type of knowledge \u2022 (causa1)word order important sequences of concepts two typeS of knowledge (s tic and semantic)
The notions of . coverage . and parsi mony are briefly sketched here for syntac tic covering through an abstract example here. \u2022 Unlike in the\u2022 case of diagnostic covering, the covers iii syntactic covering are sequences rather than sets. Consider the following descriptions \u2022 \u2022 of categories c 1 through c 5 in tenns of simpler categories (or words) w 0 through w1 0 below (sequences are indicated by being enclosed between \"<>\"):
c l : W I C 3 : <W 9 W IO W s >>
c4: <W2 w6 w3> c 5 : <w 0 w 7 >
13
", "text": "Similarities and Differences between Dia ostic Problem Solvin and Natural Lan a e Process in", "num": null }, "TABREF1": { "html": null, "type_str": "table", "content": "
[Peng and Reggia, 87; Dasigi, 88]. How
ever, for a sequence of items, the number
of irredundant covers at the next layer
grows exponentially [Dasigi, 88]. Heuristics
are needed for focu sing search in such an
ocean of covers, and semantic considera
tions seive this role. In the space of
irredundant syntactic covers, search would
be focused on \"plausible \" semantic cov
ers. 1 \u2022 Thus, the two routes of covering aid
each other by syntactic covering providing
a search space for semantic covering, and
the latter focusing further syntactic covering
at the next layer. Integration of the two
routes of covering is facilitated by attribut
ing both syntactic and semantic categories to distinguished linguistic concepts, called
op en class concepts? In general, if the
category that has just been postulated as a
cover happens to be an open class
category, .. it initiates semantic covering, thus
integrating both the routes\u2022 of covering.
Semantic covering interacts closely
withsyntacticcovering.Irredundant
14
", "text": "syntactic covering has a very nice property, namely, when complete sets of irredundant syntactic covers are considered, they are transitive across any number of layers when more than two layers of covering (e.g., as in typical parse trees) are involved", "num": null }, "TABREF3": { "html": null, "type_str": "table", "content": "", "text": "", "num": null } } } }