Citation Context

...l often leverage semantic knowledge through semantic role labeling (Pradhan, Ward, Hacioglu, Martin, & Jurafsky, 2004) and semantic similarity of terms based on large-scale resources such as WordNet (=-=Fellbaum, 1998-=-). Recent approaches seeking to increase semantic depth to improve performance (Schlaefer et al., 2007) are just reaching this level of semantic understanding, a considerable distance from general cog...

Citation Context

...and deep semantics, has seen little attention. In other fields such as linguistics, discourse psychology and philosophy there are a number of broad theories of language use (Asher & Lascarides, 2003; =-=Grice, 1975-=-; Wilson & Sperber, 2004) and narrative understanding (Graesser, Singer, & Trabasso, 1994; Ryan, 1992; Trabasso, Secco, & van den Broek, 1984) that stress the importance of world knowledge and pragmat...

Citation Context

...s (J. F. Allen & Core, 1997) are inferred over these communication events; specifically, the broad categories of answer and elaboration. The communication events in Cyc represent a speech act theory (=-=Austin, 1975-=-; Searle, 1969). These events, such as Informing, RequestingInformation and ReplyingToAQuestion, are identified through frame semantics (from terms such as “said”, “asked” and “answered”) along with t...

Citation Context

...ame There has been a significant amount of research in social psychology on how people infer internal states in others based on externally observable factors. A number of researchers, beginning with (=-=Heider, 1983-=-), have used Attribution Theory to investigate the conditions that will lead a perceiver to attribute some behavior, event or outcome to an internal disposition of the agent involved, as opposed to an...

Citation Context

... & Core, 1997) are inferred over these communication events; specifically, the broad categories of answer and elaboration. The communication events in Cyc represent a speech act theory (Austin, 1975; =-=Searle, 1969-=-). These events, such as Informing, RequestingInformation and ReplyingToAQuestion, are identified through frame semantics (from terms such as “said”, “asked” and “answered”) along with the senderOfInf...

Citation Context

...rative understanding. 6.1 Future work Allen and colleagues have reported a substantial increase in parsing speed, and a small increase in accuracy, by using output from a shallow, statistical parser (=-=Collins, 1999-=-) to bias probabilities in their deep semantic parser (Swift, Allen, & Gildea, 2004). The current implementation of QRG-CE would not likely see the same benefits, due to the strong constraints of its ...

Citation Context

...scriptions of physical phenomena in natural language text. He demonstrated that common descriptions of such phenomena effectively mapped to a frame-based representation of Qualitative Process Theory (=-=Forbus, 1984-=-). Kuehne integrated Allen’s bottom-up chart parser (J. F. Allen, 1994) with the COMLEX lexicon (Grishman et al., 1993) and subcategorization frames from Cyc. He also developed the Qualitative Reasoni...

Citation Context

...embedded knowledge about actions, events, goals and emotional reactions play a key role in comprehension. Three critical assumptions, together the principle of search after meaning first proposed by (=-=Bartlett, 1932-=-), inform this theory: 1) that readers construct meanings that 1) address their goals, 2) are globally and locally coherent and 3) explain why actions, events and states are mentioned in the text. Bas...

Citation Context

...d statistical processing methods. Subsequent work in meaningful semantic understanding of narratives turned to sophisticated logics and general-purpose inference mechanisms (Asher & Lascarides, 2003; =-=Hobbs, Stickel, Appelt, & Martin, 1990-=-; Ng & Mooney, 1990; Schubert & Hwang, 2000). However, the appeal to non-linguistic world knowledge became thinner and thinner. Simple examples used to prove formalisms are often taken without context...

Citation Context

...highly expressive sentence-level representations that delay ambiguity resolution with explicit choice sets. These representations are automatically converted into discourse representation structures (=-=Kamp & Reyle, 1993-=-) to support dynamic discourse update and query-driven reasoning at the discourse level. This allows task pragmatics to guide contextual interpretation, including disambiguation, using limited evident...

Citation Context

...oun phrases, verb phrases, prepositional phrases (PP), adjective phrases (ADJP) and adverb phrases (ADVP) in a manner similar to Allen’s textbook grammar (J. F. Allen, 1994) or other similar efforts (=-=Jurafsky & Martin, 2009-=-). Support for punctuation, sentence-level phrases (SLP) (including inversions for questions and passive voice), and coordinating conjunctions are likewise straightforward. I will not go into great de...

Citation Context

...TemporalThings, of which Situation is a specialization.70 In order to infer temporal relations, this axiomatization adopts Reichenbach’s point of reference in the representation of tense and aspect (=-=Reichenbach, 1947-=-). The point of reference combines with the point of speaking and the time of the event to fully represent the temporal aspects of a description of that event. In a sequence of event descriptions, tem...

Citation Context

...arniak and Goldman (Charniak & Goldman, 1988) proposed a logic for semantic interpretation implemented in the Wimp2 system via forward chaining in an assumption-based truth maintenance system (ATMS) (=-=deKleer, 1986-=-). This system relied on frame knowledge, expressed as facts and axioms, to make explanatory assumptions by invoking frames and seeking to fill roles. This approach can be seen as logically equivalent...

Citation Context

...wledge within the framework of this ontology both increases consistency and multiplies effort. 2.2.1.1 Events and roles in Cyc The knowledge in Cyc heavily uses Davidsonian representations of events (=-=Davidson, 2001-=-). Reified events belong to collections of such events, which generalize to the collection Event and30 through that to the collection Situation. Other features of the event are expressed using binary...

Citation Context

...nowledge required to answer such questions, giving a sense of the magnitude of the challenge. Work by Schank and his colleagues at Yale in the late 70s and early 80s (Cullingford, 1978; DeJong, 1982; =-=Schank & Ableson, 1977-=-; Wilensky, 1978) proposed specific classes of knowledge structures and showed how they applied to inferences about commonplace situations, causality, beliefs and intentions. This work was influential...

Citation Context

...ted that common descriptions of such phenomena effectively mapped to a frame-based representation of Qualitative Process Theory (Forbus, 1984). Kuehne integrated Allen’s bottom-up chart parser (J. F. =-=Allen, 1994-=-) with the COMLEX lexicon (Grishman et al., 1993) and subcategorization frames from Cyc. He also developed the Qualitative Reasoning Group Controlled English (QRG-CE) grammar used by the parser. The u...

Citation Context

...s need for rich knowledge and deep semantics, has seen little attention. In other fields such as linguistics, discourse psychology and philosophy there are a number of broad theories of language use (=-=Asher & Lascarides, 2003-=-; Grice, 1975; Wilson & Sperber, 2004) and narrative understanding (Graesser, Singer, & Trabasso, 1994; Ryan, 1992; Trabasso, Secco, & van den Broek, 1984) that stress the importance of world knowledg...

Citation Context

...n rely on general query-driven back-chaining. As is often the case, this unification provides a consistent reasoning model at the expense of per-module engineering flexibility. The TextRunner system (=-=Banko, Cafarella, Soderland, Broadhead, & Etzioni, 2007-=-) extracts relational tuples from very large scale corpora. It has been evaluated on a corpus of over 9 million web pages, demonstrating the ability to extract millions of triples with associated trut...

Citation Context

...n Propp’s terminology, the fabula is a pattern of abstract character roles and functions (actions) to be filled and performed by characters, and each sjuzhet is a different way of instantiating them (=-=Propp, 1971-=-). Barthes, a French structuralist, proposed functional units as the smallest unit of narrative. In this view a narrative is not a static artifact but rather a sequence of dynamic acts. Some are cardi...

Citation Context

...poral relations In narrative text it is critical to understand the sequencing and overlaps of the situations and events being presented. I use Cyc’s representation of Allen’s interval calculus (J. F. =-=Allen & Ferguson, 1994-=-) for specifying temporal relations among reified instances of Situation (and thus Event). These relations provide considerable semantic breadth for a wide range of configurations among arbitrary numb...

Citation Context

... on how people tell personal stories. Labov suggests a function of evaluation where the narrator follows certain conventions to identify for the hearer the most significant elements of the narrative (=-=Labov & Waletzky, 1966-=-). In this work I explore the hypothesis that an inferential task where the reader expects to infer elements such as goals, threats and outcomes as well as contrasts, parallels and commentaries can se...

Citation Context

...chniques and resources that is guided by the pragmatic concerns of the reasoning task. Sentence-level processing uses compositional frame semantics to combine knowledge-rich subcategorization frames (=-=Fillmore, 2006-=-) with a limited grammar and an optional user intervention model. This efficiently generates highly expressive sentence-level representations that delay ambiguity resolution with explicit choice sets....

Citation Context

...h as linguistics, discourse psychology and philosophy there are a number of broad theories of language use (Asher & Lascarides, 2003; Grice, 1975; Wilson & Sperber, 2004) and narrative understanding (=-=Graesser, Singer, & Trabasso, 1994-=-; Ryan, 1992; Trabasso, Secco, & van den Broek, 1984) that stress the importance of world knowledge and pragmatic context. However, they are either defined at such a high level that they provide littl...

Citation Context

...osphere (Barthes, 1977). Chatman later used the terms story and discourse, putting more emphasis on the narration process as a communicative act, in line with the pragmatics theories discussed above (=-=Chatman, 1978-=-). The central assumption of structuralist narratology is that there are identifiable structures in narrative, and that those structures are what allow the narrative to be understood. If this is the c...

Citation Context

... proven quite difficult to formalize and apply in computational models (a fact which underscores the importance of the precision and thoroughness of computational modeling). Clark and Haviland (H. H. =-=Clark & Haviland, 1977-=-) provided a notable refinement of Grice’s cooperative contract with the proposal of a given-new contract between the speaker and the listener. They suggest that a key element of understanding is reco...

Citation Context

...edge as possible. BORIS used an expectation-based conceptual parser, DYPAR, which was extended from the McELI parser (Schank & Riesbeck, 1981). This parser generated conceptual dependency (CD) forms (=-=Schank, 1972-=-) based on semantic expectations rather than syntactic constraints. BORIS remains the most in-depth implemented model of computational story understanding, able to answer questions about events, cause...

Citation Context

...re concerned with how lexical/syntactic features can be used as reasonable heuristics to identify answers. Successful approaches will often leverage semantic knowledge through semantic role labeling (=-=Pradhan, Ward, Hacioglu, Martin, & Jurafsky, 2004-=-) and semantic similarity of terms based on large-scale resources such as WordNet (Fellbaum, 1998). Recent approaches seeking to increase semantic depth to improve performance (Schlaefer et al., 2007)...

Citation Context

...gment of intention presupposes epistemic foreknowledge, but not the other way around (Shaver, 1985). Conversely, Bratman argues that epistemic foreknowledge combined with action must imply intention (=-=Bratman, 1990-=-). Acknowledging these different positions, the system makes the weaker inference that when an agent is certain of an outcome and performs or authorizes the action, it implies only some non-zero level...

Citation Context

...reasoning module to arrive at a decision. The firstprinciples reasoning module suggests decisions based on rules of moral reasoning. The analogical reasoning module uses the Structure Mapping Engine (=-=Falkenhainer, Forbus, & Gentner, 1989-=-) to compare a given scenario with previously solved decision cases to determine whether sacred values exist in the new case and suggest a course of action. Using hybridreasoning both gives the syste...

Citation Context

...matic context with real linguistic artifacts. There have been some such projects in the last few decades in the field of language understanding. Allen’s work on dialogue (J. Allen et al., 2007; J. F. =-=Allen et al., 2001-=-), Hobbs’ work on mechanical diagnosis (Hobbs, 1986) and Wilensky’s work on automating online help (Wilensky et al., 2000) integrate world knowledge and pragmatic concerns into language processing. In...

Citation Context

...ang, 1983) and baysian belief revision (Pearl, 1987). They showed that the general case is NP-hard because of the nature of the138 problems, regardless of the representation or algorithm being used (=-=Bylander, Allemang, Tanner, & Josephson, 1991-=-). Restricted classes have been found that are tractable in polynomial time, but involve assumptions unlikely to hold true in real domains (Bylander et al., 1991; Eshghi, 1992; Selman & Levesque, 1990...

Citation Context

...ving the discourse structure of a sequence of sentences. He proposes a discourse structure based on binary coherence relations between utterances, but argues that any well-formed theory (Hobbs, 1985; =-=Hovy, 1988-=-; Mann & Johnson, 1986) could apply. Hobbs argues that this approach is particularly powerful because it integrates syntactic, semantic and pragmatic constraints, not giving one type of assumption pro...

Citation Context

...arating this process from the sentencelevel composition, this approach mitigates that cost rather than placing constraints on the types of reasoning allowed. 2.2 Background 2.2.1 Cyc The Cyc project (=-=Lenat & Gupta, 1990-=-) has worked for over twenty years building a knowledge base formalizing a broad selection of common-sense background knowledge. One of the primary goals of this project is to provide knowledge that i...

Citation Context

...how specific knowledge required to answer such questions, giving a sense of the magnitude of the challenge. Work by Schank and his colleagues at Yale in the late 70s and early 80s (Cullingford, 1978; =-=DeJong, 1982-=-; Schank & Ableson, 1977; Wilensky, 1978) proposed specific classes of knowledge structures and showed how they applied to inferences about commonplace situations, causality, beliefs and intentions. T...

Citation Context

...straints). As described in section 2.3.11, utterances are represented as communication events together with a nested description of the content of the communication. Backward looking functions (J. F. =-=Allen & Core, 1997-=-) are inferred over these communication events; specifically, the broad categories of answer and elaboration. The communication events in Cyc represent a speech act theory (Austin, 1975; Searle, 1969)...

Citation Context

...that choosing the correct hypothesis out of a number of possibilities is “purely a question of economy”. Abduction was first applied in Artificial Intelligence by Pople for a medical diagnostic task (=-=Pople, 1973-=-). Abduction is well suited to model-based diagnosis where observed manifestations (the problem) must be explained. This is probably the most common use of abductive reasoning in AI, although it has a...

Citation Context

...is work served to show specific knowledge required to answer such questions, giving a sense of the magnitude of the challenge. Work by Schank and his colleagues at Yale in the late 70s and early 80s (=-=Cullingford, 1978-=-; DeJong, 1982; Schank & Ableson, 1977; Wilensky, 1978) proposed specific classes of knowledge structures and showed how they applied to inferences about commonplace situations, causality, beliefs and...

Citation Context

...icant second source of complexity in deciding which hypotheses are preferred. Bylander et al. analyzed a very general definition of abduction that covers logical formulations as well as set-covering (=-=Reggia, Nau, & Wang, 1983-=-) and baysian belief revision (Pearl, 1987). They showed that the general case is NP-hard because of the nature of the138 problems, regardless of the representation or algorithm being used (Bylander,...

Citation Context

... rather a sequence of dynamic acts. Some are cardinal, hingepoints of the plot, while others are catalysers, filling the story out with activity and diffuse concepts such as character and atmosphere (=-=Barthes, 1977-=-). Chatman later used the terms story and discourse, putting more emphasis on the narration process as a communicative act, in line with the pragmatics theories discussed above (Chatman, 1978). The ce...

Citation Context

...ingle pass over the entire corpus to classify extracted tuples as to their trustworthiness. Finally, probabilities are assigned to each trustworthy tuple based on a probabilistic model of redundancy (=-=Downey, Etzioni, & Soderland, 2005-=-). This system demonstrates scalability in automated knowledge acquisition, but at the cost of limiting consideration to clusters of word triples. 2.5.4 Logics of language Numerous logics have been pr...

Citation Context

...e understanding in the field of artificial intelligence demonstrated the crucial role of world knowledge in this inferential process. Charniak’s model of answering questions about children’s stories (=-=Charniak, 1972-=-) assumed that the interesting questions were not about the given details in a story but all the things that could be inferred from them. His work served to show specific knowledge required to answer ...

Citation Context

...n, but the power of these structured abstractions for explaining what makes a narrative coherent and controlling inferential complexity. Later work in this same tradition done by (Lehnert, 1981) and (=-=Dyer, 1983-=-) turned to abstractions of storytelling to provide yet more explanatory power. Lehnert’s plot units captured a certain predictable grammar of events in the plot of stories while Dyer’s thematic abstr...

Citation Context

...dard first-order logic (FOL) provides. Extensive work in the logic of language demonstrates that FOL is insufficient for capturing the range of ideas that can be communicated in natural language cf. (=-=Boolos, 1984-=-). Third, the expressiveness of the representation language is moot if the interpretation process cannot translate that range of ideas. The state of the art of natural language understanding is such t...

Citation Context

...resolved) 2 1 0 s3 s4 s1 s2 Scenarios Figure 23: Choice space without scoping ambiguities157 4.5 Related work Cox and Pietrzykowski present a formulation of abduction in a theorem-proving framework (=-=Cox & Pietrzykowski, 1986-=-) that defines the abducible set based on notions of causality. In this model, only basic hypotheses may be abductively assumed, those that act as a cause rather than explaining an effect in terms of ...

Citation Context

...syntactic coverage. This would not increase semantic coverage, but it would provide a more graceful failure mode where semantic fragments are still positioned within a complete parse tree. The Boxer (=-=Bos, 2005-=-) system, with the C&C Tools parser (Curran, Clark, & Bos, 2007), creates DRT-style representations over large-scale corpora. Boxer represents a significant push towards semantic depth in a statistica...

Citation Context

...aragraphs each. They are deliberately similar in an attempt to reuse as much knowledge as possible. BORIS used an expectation-based conceptual parser, DYPAR, which was extended from the McELI parser (=-=Schank & Riesbeck, 1981-=-). This parser generated conceptual dependency (CD) forms (Schank, 1972) based on semantic expectations rather than syntactic constraints. BORIS remains the most in-depth implemented model of computat...

Citation Context

...ledge in a plausible, pragmatic context with real linguistic artifacts. There have been some such projects in the last few decades in the field of language understanding. Allen’s work on dialogue (J. =-=Allen et al., 2007-=-; J. F. Allen et al., 2001), Hobbs’ work on mechanical diagnosis (Hobbs, 1986) and Wilensky’s work on automating online help (Wilensky et al., 2000) integrate world knowledge and pragmatic concerns in...

Citation Context

...multiple parse trees, but constraining the supported syntactic patterns. At the semantic level, explicit choice sets are used. This packed representation was inspired by work with Xerox’s XLE parser (=-=Riezler, King, Crouch, & Zaenen, 2003-=-). A choice set is a logical form that expresses a semantic ambiguity. Ambiguities are introduced in a number of ways, beginning with the denotations and subcategorization frames in the knowledge base...

Citation Context

...ale knowledge.81 2.5.5 Statistical parsing Statistical models of language and advances in wide-coverage grammars have made robust and efficient large-scale natural language processing a possibility (=-=Kaplan et al., 2004-=-; Matsuzaki, Miyao, & Tsujii, 2007). Such parsers can run over million and even billion-word corpora in reasonable amounts of time, identifying well-understood lexical-syntactic features. However, the...

Citation Context

...ch can be seen as logically equivalent to abductive back-chaining, with the required knowledge and axioms taking an alternative form. Later work by Charniak and Goldman used a Baysian belief network (=-=Charniak & Goldman, 1989-=-) to represent the influences between events and objects grounded in the real world. These probabilities are used to address the problem of evaluating the best explanation for the manifest text. Howev...

Citation Context

...gful semantic understanding of narratives turned to sophisticated logics and general-purpose inference mechanisms (Asher & Lascarides, 2003; Hobbs, Stickel, Appelt, & Martin, 1990; Ng & Mooney, 1990; =-=Schubert & Hwang, 2000-=-). However, the appeal to non-linguistic world knowledge became thinner and thinner. Simple examples used to prove formalisms are often taken without context and where necessary world knowledge is inv...

Citation Context

...on (divided into type and relation concretion) as a classof inference (Norvig, 1987). Later work moved away from marker passing to general logic programming as a framework. 158 Charniak and Goldman (=-=Charniak & Goldman, 1988-=-) proposed a logic for semantic interpretation implemented in the Wimp2 system via forward chaining in an assumption-based truth maintenance system (ATMS) (deKleer, 1986). This system relied on frame ...

Citation Context

...theses are preferred. Bylander et al. analyzed a very general definition of abduction that covers logical formulations as well as set-covering (Reggia, Nau, & Wang, 1983) and baysian belief revision (=-=Pearl, 1987-=-). They showed that the general case is NP-hard because of the nature of the138 problems, regardless of the representation or algorithm being used (Bylander, Allemang, Tanner, & Josephson, 1991). Res...

Citation Context

...mar used by the parser. The use of a limited-syntax language to factor out parsing difficulties was inspired by both CMU’s KANT project (Nyberg et al., 2002) and Boeing’s controlled language work (P. =-=Clark, Harrison, Jenkins, Thompson, & Wojcik, 2005-=-). Like these simplified languages, QRG-CE restricts grammar but does not a priori restrict the vocabulary. This enables most extensions to be made by adding vocabulary rather than changing the gramma...

Citation Context

...course structure of a sequence of sentences. He proposes a discourse structure based on binary coherence relations between utterances, but argues that any well-formed theory (Hobbs, 1985; Hovy, 1988; =-=Mann & Johnson, 1986-=-) could apply. Hobbs argues that this approach is particularly powerful because it integrates syntactic, semantic and pragmatic constraints, not giving one type of assumption prominence over another. ...

Citation Context

...lizing novel plans used a similar approach to Dyer, combining schema recognition with plan knowledge to process three shorter stories as well as three stories of three (single clause) sentences each (=-=Mooney, 1988-=-). His evaluation of inferences was more limited than Dyer’s, being concerned only with planning knowledge. Following work by Goldman (Goldman, 1991) and Ng (Ng, 1992) on plan recognition in narrative...

Citation Context

...uding discussion. 4.1 Background 4.1.1 Abduction Abductive reasoning is a nonmonotonic form of reasoning that searches for explanations for observed manifestations. It was first formalized by Peirce (=-=Peirce, 1955-=-) who concluded that if the presence of a “surprising fact” C could be explained by the presence of a fact A, this is reason to suspect that A is true. He also argues that the hypothesis that A is tru...

Citation Context

... 1979; Thorndyke, 1977; van Dijk, 1975). These theories were concerned with how human readers store narratives in memory, and how that storage mechanism impacts recall. They were strongly criticized (=-=Black & Wilensky, 1979-=-), defended against those criticisms (Mandler & Johnson, 1980; Rumelhart, 1980) and then232 criticized again (Garnham, 1983). Black and Wilensky took issue with the ability of a formal grammar to ade...

Citation Context

...a number of higher-order features, including quantification over predicates, functions, and sentences, and the ability for predicates to take predicates as values, including themselves in some cases (=-=Ramachandran, Reagan, & Goolsbey, 2005-=-). Higher-order predicates are used for logical relations (e.g. and, or, implies), quantification (e.g. forAll, thereExists) and modality (e.g. willBe, possible). Every assertion in CycL occurs in the...

Citation Context

...ree stories of three (single clause) sentences each (Mooney, 1988). His evaluation of inferences was more limited than Dyer’s, being concerned only with planning knowledge. Following work by Goldman (=-=Goldman, 1991-=-) and Ng (Ng, 1992) on plan recognition in narrative text moved away from schema recognition and used only simple one to four sentence stories. Each sentence presented a single state or event regardin...

Citation Context

...l-purpose concretion engine, Norvig proposed a general-purpose inferential system based on marker passing that included concretion (divided into type and relation concretion) as a classof inference (=-=Norvig, 1987-=-). Later work moved away from marker passing to general logic programming as a framework. 158 Charniak and Goldman (Charniak & Goldman, 1988) proposed a logic for semantic interpretation implemented i...

Citation Context

...een some such projects in the last few decades in the field of language understanding. Allen’s work on dialogue (J. Allen et al., 2007; J. F. Allen et al., 2001), Hobbs’ work on mechanical diagnosis (=-=Hobbs, 1986-=-) and Wilensky’s work on automating online help (Wilensky et al., 2000) integrate world knowledge and pragmatic concerns into language processing. In doing so, they point the field towards using the w...

Citation Context

...lations In narrative text it is critical to understand the sequencing and overlaps of the situations and events being presented. I use Cyc’s representation of Allen’s interval calculus (J. F. Allen & =-=Ferguson, 1994-=-) for specifying temporal relations among reified instances of Situation (and thus Event). These relations provide considerable semantic breadth for a wide range of configurations among arbitrary numb...

Citation Context

... readers store narratives in memory, and how that storage mechanism impacts recall. They were strongly criticized (Black & Wilensky, 1979), defended against those criticisms (Mandler & Johnson, 1980; =-=Rumelhart, 1980-=-) and then232 criticized again (Garnham, 1983). Black and Wilensky took issue with the ability of a formal grammar to adequately capture the diverse forms found in narrative, and argued that a syntac...

Citation Context

....5 Statistical parsing Statistical models of language and advances in wide-coverage grammars have made robust and efficient large-scale natural language processing a possibility (Kaplan et al., 2004; =-=Matsuzaki, Miyao, & Tsujii, 2007-=-). Such parsers can run over million and even billion-word corpora in reasonable amounts of time, identifying well-understood lexical-syntactic features. However, the translation from such surface fea...

Citation Context

... of the narration. Shklovsky and Tomashevski identified the fabula as a story as it would occur as a literal sequence of events, contrasted with the sjuzhet, a narrative that communicates that story (=-=Bertens, 2008-=-). Propp studied a hundred Russian folktales and concluded that they were all variations of the same story, different sjuzhets for the same fabula. In Propp’s terminology, the fabula is a pattern of a...

Citation Context

...narrative understanding. 5.1.5 Inferring coherence and relevance Work in computational models of narrative understanding has focused on inferring coherent interpretations. The early work by Charniak (=-=Charniak, 1977-=-) and Schank’s group at Yale (Cullingford, 1978; Schank & Ableson, 1977; Wilensky, 1978) used frames, scripts, plans and other knowledge structures to identify events in a narrative as parts of larger...

Citation Context

...miautomating it, tailorability is reduced, and the plausibility of the simulation results is increased. EA NLU has been used in several cognitive modeling experiments including moral decision making (=-=Dehghani, Tomai, Forbus, & Klenk, 2008-=-), conceptual change (Friedman & Forbus, 2008) and blame attribution (Tomai & Forbus, 2008). In this chapter I describe how EA NLU has been used in three cognitive modeling experiments. I provide evid...

Citation Context

...uent work in meaningful semantic understanding of narratives turned to sophisticated logics and general-purpose inference mechanisms (Asher & Lascarides, 2003; Hobbs, Stickel, Appelt, & Martin, 1990; =-=Ng & Mooney, 1990-=-; Schubert & Hwang, 2000). However, the appeal to non-linguistic world knowledge became thinner and thinner. Simple examples used to prove formalisms are often taken without context and where necessar...

Citation Context

...nations. Global cost metrics, whether hand represented or based on Baysian probability theory, have been criticized as inflexible because they do not use the particular context of the interpretation (=-=Norvig & Wilensky, 1990-=-). This is exacerbated by the fact that interesting narratives are often pointedly about uncommon or improbable occurrences. Coherence within the explanation, such as measured by factoring redundancy ...

Citation Context

...e construction-integration theory (CI) (Kinsch, 1998) of reading has seen computational implementation (Mross & Roberts, 1992) and been evaluated as an explanatory theory for133 bridging inferences (=-=Schmalhofer, McDaniel, & Keefe, 2002-=-). CI posits two separate, repeating processing phases: construction and integration. The construction phase takes a new clause and instantiates or activates knowledge units as nodes in a multi-layer ...

Citation Context

... 3.1, consisting of roughly 86000 dictionary entries. For additional details, see the reference manual (Macleod, Grishman, & Meyers, 1998). 2.2.3 EA NLU The EA NLU system was originally developed by (=-=Kuehne, 2004-=-) to explore understanding descriptions of physical phenomena in natural language text. He demonstrated that common descriptions of such phenomena effectively mapped to a frame-based representation of...

Citation Context

...gy and philosophy there are a number of broad theories of language use (Asher & Lascarides, 2003; Grice, 1975; Wilson & Sperber, 2004) and narrative understanding (Graesser, Singer, & Trabasso, 1994; =-=Ryan, 1992-=-; Trabasso, Secco, & van den Broek, 1984) that stress the importance of world knowledge and pragmatic context. However, they are either defined at such a high level that they provide little specific f...

Citation Context

... concerned with how human readers store narratives in memory, and how that storage mechanism impacts recall. They were strongly criticized (Black & Wilensky, 1979), defended against those criticisms (=-=Mandler & Johnson, 1980-=-; Rumelhart, 1980) and then232 criticized again (Garnham, 1983). Black and Wilensky took issue with the ability of a formal grammar to adequately capture the diverse forms found in narrative, and arg...

Citation Context

...artin, & Jurafsky, 2004) and semantic similarity of terms based on large-scale resources such as WordNet (Fellbaum, 1998). Recent approaches seeking to increase semantic depth to improve performance (=-=Schlaefer et al., 2007-=-) are just reaching this level of semantic understanding, a considerable distance from general cognitive modeling. Simlilar NLP techniques are used in the Text Analysis Conference (TAC) summarization ...

Citation Context

...ingle clause) sentences each (Mooney, 1988). His evaluation of inferences was more limited than Dyer’s, being concerned only with planning knowledge. Following work by Goldman (Goldman, 1991) and Ng (=-=Ng, 1992-=-) on plan recognition in narrative text moved away from schema recognition and used only simple one to four sentence stories. Each sentence presented a single state or event regarding a small set of p...

Citation Context

...t storage mechanism impacts recall. They were strongly criticized (Black & Wilensky, 1979), defended against those criticisms (Mandler & Johnson, 1980; Rumelhart, 1980) and then232 criticized again (=-=Garnham, 1983-=-). Black and Wilensky took issue with the ability of a formal grammar to adequately capture the diverse forms found in narrative, and argued that a syntactic characterization of story structure is red...

Citation Context

...tences to proving the discourse structure of a sequence of sentences. He proposes a discourse structure based on binary coherence relations between utterances, but argues that any well-formed theory (=-=Hobbs, 1985-=-; Hovy, 1988; Mann & Johnson, 1986) could apply. Hobbs argues that this approach is particularly powerful because it integrates syntactic, semantic and pragmatic constraints, not giving one type of as...

Citation Context

................................................ 61 Figure 4: Moral decision making scenario from Ritov and Baron 1999 ...................................... 87 Figure 5: Corporate Program Scenario from =-=Mao 2006-=- ............................................................... 91 Figure 6: DRS for “A convoy of trucks is transporting food to a refugee camp during a famine in Africa.” ..............................

Citation Context

...ns, preferring inactions to actions. For a given scenario a set of rules are applied to decide whether the case includes sacred values or not. An orders of magnitude reasoning module based on ROM(R) (=-=Dauge, 1993-=-) then calculates the relationship between the utility of each choice. Using the outcome of the orders of magnitude reasoning module, MoralDM utilizes a hybrid reasoning approach consisting of a first...

Citation Context

...nderstanding was also quite common in psychology in the 1970s, with theories of story grammar being one of the most prominent. Lakoff reformulated Propp’s morphology as a grammar using rewrite rules (=-=Lakoff, 1972-=-), and Rumelhart (Rumelhart, 1975) proposed a general grammar aimed at all stories. This was followed by numerous general grammars (Mandler & Johnson, 1977; Stein & Glenn, 1979; Thorndyke, 1977; van D...

Citation Context

...entions. This work was influential, but the use of rich knowledge in language understanding largely disappeared in the research community due to concerns about the lack of robustness and scalability (=-=Lehnert, 1994-=-) as well as the high knowledge engineering cost for applicable world knowledge. The statistical revolution in natural language research shifted the focus almost entirely towards shallow, well defined...

Citation Context

...d perhaps even more so) being able to give an account of its “message”, describe what (more or less) general subject or truth it illustrates, specify what “it is getting at”, put forth its “point”.” (=-=Prince, 1983-=-) Prince goes on to describe the need to better understand the pragmatic concerns that lead to inferring this point, acknowledging that narrative pragmatics is “…anything and everything that seems to ...

Citation Context

... th century and enjoyed immense popularity, solidifying the fable as a distinct genre in English literature. These fables are attributed to the Greek slave Aesop, believed to have lived around 550BC (=-=Lewis, 1996-=-). Aesop’s fables exemplify the use of narrative to convey a specific point. The inclusion of a moral, a one-sentence maxim separate from the story proper, strongly indicates authorial intent. The rea...

Citation Context

...llection of assertion acts (Searle, 1975). It has been argued that this is not sufficient to account for either the obligations that surround a narration or the illocutionary force that it can exert (=-=Nair, 2003-=-). Being told a164 narrative is not merely a stand-in for directly observing the same set of events in the real world. Even a plain narration that lacks commentary or privileged insight must choose a...

Citation Context

...ommon use of abductive reasoning in AI, although it has also seen use in, among other areas, plan recognition, planning, case-based137 reasoning, learning, vision and natural language understanding (=-=Paul, 2000-=-). Pople formulated abduction as linear resolution in a theorem-proving framework, stating that for a logical theory Τ and a formula ω which is the fact to be explained, a cause φ of ω in Τ is determi...

Citation Context

...marization challenge. Features such as named entities and parts of speech are combined with WordNet semantic distance metrics to score the importance of sentences within a larger text for distilling (=-=Bawakid & Oussalah, 2008-=-). This area has developed many practical information retrieval and text manipulation techniques. It has also served to better define what can and cannot be accomplished in a knowledge-poor environmen...

Citation Context

... large corpora. By contrast, this work is concerned with representing the content of a narrative as a whole, requiring more complex relations between those true facts. The Cycorp TextLearner project (=-=Curits et al., 2006-=-) leverages the knowledge in the Cyc knowledge base to generate an information-rich model of a document which can then be used as a guide for learning. TextLearner is concerned with learning both docu...

Citation Context

..., Allemang, Tanner, & Josephson, 1991). Restricted classes have been found that are tractable in polynomial time, but involve assumptions unlikely to hold true in real domains (Bylander et al., 1991; =-=Eshghi, 1992-=-; Selman & Levesque, 1990). 4.1.2 Abduction and Natural Language Understanding Early work in natural language understanding produced the insight that interpretation is a process of forming a reasonabl...

Citation Context

...son, Lear, & Andrusiak, 1992; Suh & Trabasso, 1993). These predicted inferences could form the basis of an interesting cognitive model for reading, one that could be tested in EA NLU. Experiments in (=-=Kim, 1999-=-) tested the hypothesis that the presence of causal bridging inferences in a story influences the perception of interestingness. Two versions of the same stories were constructed, one with five causal...

Citation Context

...or maintaining a coherent model. This is particularly interesting as a model of inference because it defines the implications of specific contextual factors. The construction-integration theory (CI) (=-=Kinsch, 1998-=-) of reading has seen computational implementation (Mross & Roberts, 1992) and been evaluated as an explanatory theory for133 bridging inferences (Schmalhofer, McDaniel, & Keefe, 2002). CI posits two...

Citation Context

...ledge and expectation, but the power of these structured abstractions for explaining what makes a narrative coherent and controlling inferential complexity. Later work in this same tradition done by (=-=Lehnert, 1981-=-) and (Dyer, 1983) turned to abstractions of storytelling to provide yet more explanatory power. Lehnert’s plot units captured a certain predictable grammar of events in the plot of stories while Dyer...

Citation Context

...rongly limits the range of narratives that can be expressed. Work on memory and recall of narratives analyzed short stories to create hierarchical story grammars, such as that of Mandler and Johnson (=-=Mandler & Johnson, 1977-=-). The grammar was used to manually create story schemata, hypothesized as representative of human memory organization, to make predictions about what parts of stories would be remembered and recalled...

Citation Context

...he Qualitative Reasoning Group Controlled English (QRG-CE) grammar used by the parser. The use of a limited-syntax language to factor out parsing difficulties was inspired by both CMU’s KANT project (=-=Nyberg et al., 2002-=-) and Boeing’s controlled language work (P. Clark, Harrison, Jenkins, Thompson, & Wojcik, 2005). Like these simplified languages, QRG-CE restricts grammar but does not a priori restrict the vocabulary...