Regier, T. P. (1992). The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. PhD thesis, University of California at Berkeley.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of event types. Borchardt s system receives support and contact information as input, in contrast to my system, which calculates such information using counterfactual simulation. Herskovits (1986) describes an unimplemented theory of the truth conditions underlying English spatial prepositions. Regier (1992) describes an implemented system that can learn the truth conditions on the use of spatial terms in a variety of languages in a languageindependent fashion. Funt (1980) describes a counterfactual simulator for determining support relationship between objects in a static image. His system operates ....

Regier, T. P. (1992). The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. PhD thesis, University of California at Berkeley.

....representation of a particular relation solely on the basis of positive examples, leading to dramatic over generalisation during testing. Subsequently, training was adapted to employ positive examples of an antonymic relation for example below in the case of above as implicit negative evidence [13] for the desired category, and overgeneralisation was reduced to acceptable levels. Variations in sequence length are considered by employing movies of 3; 5 and 7 frames for static relations, and 5 and 7 frames for dynamic relations. Thus, when variations in trajector and landmark position are ....

Regier, T., \The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization". PhD Dissertation, University of California, Berkeley, (1992).

....and grouping are less obviously so. Calibration and grounding refer to the agent s ability to coordinate perception and motion and (in our view) explicit beliefs about the agent s body in relation to near space. Some related work by others on calibration, grounding, and self models includes [1, 9, 15, 5, 7]. Confusion and contradiction can easily arise in near space (due to noisy or conflicting data from multiple modalities, or from defaults gone wrong) and the recognizing of such is a crucial step in setting one s world view right again; this is where symbolically explicit calibration can enter ....

T. Regier. The acquisition of lexical semantics for spatial terms: A connectionist model of perceptual categorization. PhD thesis, University of California at Berkeley, 1992.

.... by (Langacker, 1987) elementary spatial concepts (such as the English above) are characterised by locative relations between a potentially mobile object called the trajector (TR) and a static reference object called the landmark (LM) Previous computational investigations of this problem (Regier, 1992), have relied upon highly structured feature detection systems and the abstraction of object identification issues into the input data. While highly successful on their own terms, systems of this nature are not readily generalisable to problems involving more sophisticated (especially cluttered) ....

....in the exclusion of explicit negative evidence (see for example (Chomsky, 1965) Thus only positive instances of a given concept may be presented during training, but the system may receive negative examples during normal operation. 2. 1 A Semantic Sub Task The L 0 sub task examined by (Regier, 1992) requires that the model system acquire perceptually grounded semantics of natural language spatial terms . Each lexeme describes a locative relationship between a special (potentially mobile) object known as the trajector (TR) and a static reference object known as the landmark (LM) Langacker, ....

[Article contains additional citation context not shown here]

Regier, T. 1992. The Acquisition of Lexical Semantics for Spatial terms: A Connectionist Model of Perceptual Categorization. PhD Dissertation, University of California, Berkeley.

....events correspond roughly to the verb vocabulary in language. The overall goal of this research is to ground the lexical semantics of verbs in visual perception. A number of reported systems can classify event occurrences from video or simulated video, among them, Yamoto, Ohya, Ishii (1992) Regier (1992), Pinhanez Bobick (1995) Starner (1995) Siskind Morris (1996) Bailey et al. 1998) and Bobick Ivanov (1998) While they differ in their details, by and large, these system classify event occurrences by their motion profile. For example, a pick up event is described as a sequence of two ....

Regier, T. P. 1992. The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. Ph.D. Dissertation, University of California at Berkeley.

....cannot be adequately explained in terms of other words (Harnad, 1990) The core of their meaning is rather directly associated with visual perception. This means that to a significant extent, the acquisition of spatial relation concepts has to be independent of language (Choi and Bowerman, 1991, Regier, 1992). This, in turn, implies that humans have some innate capability to automatically form spatial relation concepts, by noticing regularities in their visual input (unsupervised learning) When words are present (supervised learning) they help refine those concepts to their intended meaning within ....

....justifies our assumptions of preprocessing. There s just too much going on in what an infant perceives, for it to attain spatial categories just by hearing the correct words in the correct contexts, without any inborn attentive predispositions. How far these extend, is we think an open question. Regier (Regier, 1992) has achieved remarkable results in the field of spatial concept acquisition, both static and dynamic, across languages, by assuming a considerable amount of preprocessing, which is all hardwired in his architecture and precedes learning. We think we have shown that with relatively simple and ....

Regier, T. 1992. The Acquisition of Lexical Semantics for Spatial Terms : A Connectionist Model of Perceptual Categorization. Ph.D. diss., Dept. of Computer Science, University of California at Berkeley.

....without the benefit of explicit negative evidence, and as a result of its structure and operation, gives rise to a number of linguistic predictions concerning which semantic features one may or may not expect to find in spatial lexemes in the world s languages. We refer the interested reader to (Regier 1992; Regier 1993) for further details. 4 Current and Planned Efforts The problem of learning and representing spatial relations terms in the world s language is extraordinarily difficult, and Regier s model is only a beginning. It is limited to 2dimensional representations, and to artificial scenes ....

Regier, Terry, 1992. The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. Computer Science Division, EECS Department, University of California at Berkeley dissertation. available as Technical Report TR-92-062, International Computer Science Institute, Berkeley.

....also suggested how some aspects of language typically associated with compositionality could be modeled, without there being a strictly compositional semantics . This is done through a series of experiments, using modifications of Terry Regier s connectionist system for learning spatial relations [Regier, 1992] which constitutes a part of the L 0 project concerned with associating descriptions in an arbitrary language with an analog environment, sequences of) pictures of simple 2 dimensional scenes. The emphasis is above all on the English preposition over , famous for its polysemy, and analyzed in ....

....a number of distinct, but related meanings 3 . So to explain polysemy, one would have to give answers to questions such as: What is a sense meaning of a lexical item (word) In which ways can distinct senses be related Why should the words of human languages display polysemy at all 3 [Regier, 1992], for example, introduces his section on polysemy with the sentence Polysemy is a linguistic phenomenon whereby a single word has a cluster of distinct but related senses. p.21) Regier s system hover climb fly over go be hover over climb over fly over go over be over Figure 1: A ....

[Article contains additional citation context not shown here]

Terry Regier, The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization, PhD thesis, University of California at Berkeley, December 1992, (to appear).

.... procedures by outlining a method that used distal teacher information, that is, 2 In the domain of language learning, it has been suggested that the poor availability of negative feedback makes the learning of grammars for non finite languages impossible (Gold, 1967) See MacWhinney (1993) and Regier (1992) for a discussion of possible ways to avoid this conclusion. information not directly related to the outputs of the network. They provided the example of a basketball player learning to shoot baskets. To successfully propel the ball through the hoop a player must issue a complex sequence of ....

Regier, T. (1992). The acquisition of lexical semantics for spatial terms: A connectionist model of perceptual categorization. Technical Report TR-92-062, International Computer Science Institute, Berkeley, CA.

.... of the proposal was to highlight certain fundamental problems with traditional cognitive science theories, including issue such as dependence on the underlying conceptual system, grounding of meaning and categorization in perception, and others which are explored in recent and ongoing research (Regier 1992; Feldman et al. 1994) For our purposes we can abstract a (much simpler) subproblem from this interdisciplinary task: given pairs of sentences and associated idealized semantics (e.g. in first order logic formulae) construct an adequate formal description of the relation between these two for ....

REGIER, TERRY, 1992. The acquisition of lexical semantics for spatial terms: A connectionist model of perceptual categorization. Berkeley, Ca.: Computer Science Division, University of California dissertation.

.... work by Feldman and Ballard, who provided a general framework of structured connectionism [15] This framework was extended for natural language processing in many different directions, for instance for parsing [12, 13, 46] language acquisition [14, 25] explanation [8] and semantic processing [26, 47]. More recent work along these lines focuses on the so called NTL, neural theory of language which attempts to bridge the large gap between neurons and cognitive behavior [16, 49] The NTL framework is challenging since it tries to study neural processing mechanisms for high conceptual cognitive ....

T. Regier. The acquisition of lexical semantics for spatial terms: a connectionist model of perceptual categorization. Technical Report Technical Report, International Computer Science Institute, 1992.

....rules. In each case, each word consisted of two morphemes, a root and a single tense inflection, marking the present or past . Examples of each rule: ffl Suffix: present vibuni, past vibuna ffl Prefix: present ivibun, past avibun ffl Infix: present vikbun, past vinbun 10 See Regier (1992) and Gasser Smith (1993) for two approaches to negative evidence and mutual exclusivity when single categories for multiple inputs are frequent enough to present a problem. ffl Circumfix: present ivibuni, past avibuna ffl Mutation: present vibun, past vib un ffl Deletion: present vibun, ....

Regier, T. (1992). The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. Ph.D. thesis, University of California, Berkeley.

.... is prespecified to suit a particular task (Feldman et al. 1988) Some examples of structured connectionist models are: connectionist knowledge representation systems like shruti (Shastri and Ajjanagadde, 1993; Mani and Shastri, 1993) csn (Shastri, 1988) and robin (Lange and Dyer, 1989) Terry Regier s (1992) system which learns the lexical semantics of spatial terms; Nigel Goddard s (1992) architecture for recognizing moving light displays; systems for visual motion understanding (Olson, 1989) visual attention, and visual search (Ahmad, 1991) and adjective noun interpretation (Webber, 1989) In ....

Regier, T. (1992). The acquisition of lexical semantics for spatial terms: A connectionist model of perceptual categorization. Technical Report TR-92-062, International Computer Science Institute.

....provide a grounding for concrete object terms by associating the terms with sensory stimuli. A similar approach using machines seems to be an appropriate solution to the part of the symbol grounding problem that deals with concrete concepts. Several workers in Artificial Neural Networks (ANNs) [6,7,8] in more traditional Artificial Intelligence [9,10] and in hybrid systems using both technologies [11] have approached this problem in principle, although not with large, live video visual inputs in an interactive way, as this paper does. Furthermore, this work represents a step ....

Regier, T., "The Acquisition Of Lexical Semantics For Spatial Terms: A Connectionist Model Of Perceptual Categorization", PhD Thesis, available as Tech. Report: TR-92-062, International Computer Science Institute, Berkeley, CA, 1992.

.... Structure Since Harnad s important paper on the need to ground symbols in perception or action (Harnad, 1990) there have been various attempts to train systems to categorize visual inputs in terms of noun or noun like categories or simple one or two place relations (Dorffner, 1990; Nenov, 1991; Regier, 1992). While the progress made by these researchers should not be minimized, grounded categories by themselves are only the beginning. One of the hallmarks of human concepts is their structure. On the traditional view, symbols combine to form an effectively unlimited set of possible symbol structures, ....

Regier, T. (1992). The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. Ph.D. thesis, University of California, Berkeley.

....sentences with visual stimuli. Children who have learned nouns like ball and boy can begin to learn the meaning of kick, by observing a boy kicking a ball when they hear the sentence the boy kicks the ball. One relatively simple system developed to simulate this phenomena was developed by Regier[19]. Regier developed a connectionist network which learned the meaning of phrases by associating them with a visual stimuli consisting of two objects (one stationary, one moving) in a 2 D microworld. This system is a part of the research in the L 0 project[6] which has the long term goal of ....

T. Regier. The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. PhD thesis, University of California at Berkeley, 1992.

....and grouping are less obviously so. Calibration and grounding refer to the agent s ability to coordinate perception and motion and (in our view) explicit beliefs about the agent s body in relation to near space. Some related work by others on calibration, grounding, and self models includes [1, 9, 15, 5, 7]. Confusion and contradiction can easily arise in near space (due to noisy or conflicting data from multiple modalities, or from defaults gone wrong) and the recognizing of such is a crucial step in setting one s world view right again; this is where symbolically explicit calibration can enter ....

T. Regier. The acquisition of lexical semantics for spatial terms: A connectionist model of perceptual categorization. PhD thesis, University of California at Berkeley, 1992.

....and for the existence of cells with circularly symmetric center surround receptive #elds #Ku#er 1953#. These feature maps are used in the detection of non orientational spatial features such as contact and inclusion. The interested reader may #nd further details regarding these structures in #Regier 1992#. 5.1 Motion Bu#ers: Tripartite Trajectory Representation While the evidence for orientation sensitive cells, topographic maps, and cells with circularly symmetric center surround receptive #elds comes from neurobiology, the motion bu#ers are motivated by psychophysical and linguistic evidence. ....

....independent of the time step during which it occurred. These predicted disadvantages are borne out by empirical testing, which has shown that the motion bu#er approach outperforms simple recurrentnetworks when applied to the task of learning closed class categorizations of spatial relations #Regier 1992#. Of course, the motion bu#er approach, with its deliberate disregard for intermediate sequentialityof events, would be essentially useless for grammatical induction or other tasks in which temporally intermediate sequentiality is of the essence. 5.2 Learning without Explicit Negative Evidence ....

[Article contains additional citation context not shown here]

Regier, Terry, 1992. The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. Computer Science Division, EECS Department, University of California at Berkeley dissertation. available as Technical Report TR-92-062, International Computer Science Institute, Berkeley.

....and for the existence of cells with circularly symmetric center surround receptive fields (Kuffler 1953) These feature maps are used in the detection of non orientational spatial features such as contact and inclusion. The interested reader may find further details regarding these structures in (Regier 1992). 5.1 Motion Buffers: Tripartite Trajectory Representation While the evidence for orientation sensitive cells, topographic maps, and cells with circularly symmetric center surround receptive fields comes from neurobiology, the motion buffers are motivated by psychophysical and linguistic ....

....independent of the time step during which it occurred. These predicted disadvantages are borne out by empirical testing, which has shown that the motion buffer approach outperforms simple recurrent networks when applied to the task of learning closed class categorizations of spatial relations (Regier 1992). Of course, the motion buffer approach, with its deliberate disregard for intermediate sequentiality of events, would be essentially useless for grammatical induction or other tasks in which temporally intermediate sequentiality is of the essence. 5.2 Learning without Explicit Negative Evidence ....

[Article contains additional citation context not shown here]

Regier, Terry, 1992. The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. Computer Science Division, EECS Department, University of California at Berkeley dissertation. available as Technical Report TR-92-062, International Computer Science Institute, Berkeley.

....learn without the bene#t of explicit negative evidence, and as a result of its structure and operation, gives rise to a number of linguistic predictions concerning whichsemantic features one mayormay not expect to #nd in spatial lexemes in the world s languages. We refer the interested reader to #Regier 1992; Regier 1993# for further details. 4 Current and Planned E#orts The problem of learning and representing spatial relations terms in the world s language is extraordinarily di#cult, and Regier s model is only a beginning. It is limited to 2dimensional representations, and to arti#cial scenes ....

Regier, Terry, 1992. The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. Computer Science Division, EECS Department, University of California at Berkeley dissertation. available as Technical Report TR-92-062, International Computer Science Institute, Berkeley.

No context found.

Terry Regier. The Acquisition of Lexical Semantics for Spatial Terms: A Connectionist Model of Perceptual Categorization. PhD thesis, ICSI, Berkeley, 1992.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC