The five-year ARPA-funded speech project that began at that time made understanding, rather than The transduction from speech to meaning must be mediated recognition, the primary research goal. It was felt that a by a variety of components that utilize diverse system's ability to respond intelligently to speech was a knowledge sources (KSs) because the speech signal encodes, more meaningful criterion for the evaluation of speech in a highly compressed and integrated fashion, systems. In addition, it was believed that the speech signal many different types of information relevant to the recovery was an impoverished source of information, and of meaning. This knowledge-based approach contrasts knowledge of the context of an utterance was essential for with that taken in whole-word template-matching systems; its successful recognition and interpretation. Speech-recognition variability in the pronunciation of words in connected systems based on dynamic programming, pattern-matching speech is no longer seen as a hindrance to pattern techniques have been developed for utterances matching but rather as an important source of information, that consist solely of isolated words chosen from a eg, concerning the location of word boundaries small vocabulary, and to a lesser extent, the same techniques (Church, 1983) or of contextually important (stressed) information have been extended to connected sequences of in the utterance. Figure 1 illustrates one possi-words Rabiner and Levinson (1981).

Step 6 is a goal-assertion the input, another algorithm might result. Thus one could resolution that functions similarly to the goal-goal resolution break a into a[1],..., a [length(a)/2] and a [length(a)/ above. The final synthesized program is: 2 1],..., a[length(a)] and find an algorithm that recursively calls f on both the first and second halves of its f(x) if x NIL then 0 else car(x) f(cdr(x)).

Our descriptive vocabulary may still In this article I propose to describe an area of artificial contain proper names as modifiers but the explanatory intelligence (Al) research that I and several colleagues vocabulary now involves the impersonal qualities of an have been enaged in for a number of years.

Psychological work on memory, in particular by Bartlett (1932), has led the conclusion that people faced with a new situation use large amounts of highly structured knowledge acquired from previous experience. Bartlett used the word schema to refer to this phenomenon. Minsky (1975), his famous paper, proposed the notion of a frame as a fundamental structure used in natural language understanding, as well as in scene analysis. I will use the former term in the rest of this chapter, in spite of its general connotation. The main thesis defended by Bartlett was that the phenomena of memorization and remembering are both constructive and selective. The hypothesis has more recently been revived by psychologists working on discourse structure (Collins, 1978; Bransford and Franks, 1971; Kintsch, 1976). Various experiments performed on subjects who were told stories and then asked to describe what they remembered showed that people not only forget facts but add some. Moreover, they are unable to distinguish between what they have actually heard and what they have inferred. People hearing a story make assumptions, which they might revise or refine as more information comes in, either confirmatory or contradictory. Making such assumptions entails building (or retrieving) models of the expected text contents. A corollary of this process is that if the story adequately fits the model people have in mind, the story will be understood more easily. This chal)ter is based on a technical memo (HPP-79-25) from the Heuristic Programming lh( iect, l)cparmlent of Computer Science, Stanford University.

The ability how they can be combined. Thus the user would be to program in natural language instead of traditional taxed more heavily with a natural language system programming languages would enable people to use than with a traditional system. A second argument familiar constructs in expressing their requests, thus against natural language programming relates to its making machines accessible to a wider user group.

IDijkstra 19591 Dijkstra, E., NA Note on Two Problems in Connection with Graphs," Numeri.scht'

This article is an overview of the "SP theory of intelligence". The theory aims to simplify and integrate concepts across artificial intelligence, mainstream computing and human perception and cognition, with information compression as a unifying theme. It is conceived as a brain-like system that receives 'New' information and stores some or all of it in compressed form as 'Old' information. It is realised in the form of a computer model -- a first version of the SP machine. The concept of "multiple alignment" is a powerful central idea. Using heuristic techniques, the system builds multiple alignments that are 'good' in terms of information compression. For each multiple alignment, probabilities may be calculated. These provide the basis for calculating the probabilities of inferences. The system learns new structures from partial matches between patterns. Using heuristic techniques, the system searches for sets of structures that are 'good' in terms of information compression. These are normally ones that people judge to be 'natural', in accordance with the 'DONSVIC' principle -- the discovery of natural structures via information compression. The SP theory may be applied in several areas including 'computing', aspects of mathematics and logic, representation of knowledge, natural language processing, pattern recognition, several kinds of reasoning, information storage and retrieval, planning and problem solving, information compression, neuroscience, and human perception and cognition. Examples include the parsing and production of language including discontinuous dependencies in syntax, pattern recognition at multiple levels of abstraction and its integration with part-whole relations, nonmonotonic reasoning and reasoning with default values, reasoning in Bayesian networks including 'explaining away', causal diagnosis, and the solving of a geometric analogy problem.

Formalisms for specifying statistical models, such as probabilistic-programming languages, typically consist of two components: a specification of a stochastic process (the prior), and a specification of observations that restrict the probability space to a conditional subspace (the posterior). Use cases of such formalisms include the development of algorithms in machine learning and artificial intelligence. We propose and investigate a declarative framework for specifying statistical models on top of a database, through an appropriate extension of Datalog. By virtue of extending Datalog, our framework offers a natural integration with the database, and has a robust declarative semantics. Our Datalog extension provides convenient mechanisms to include numerical probability functions; in particular, conclusions of rules may contain values drawn from such functions. The semantics of a program is a probability distribution over the possible outcomes of the input database with respect to the program; these outcomes are minimal solutions with respect to a related program with existentially quantified variables in conclusions. Observations are naturally incorporated by means of integrity constraints over the extensional and intensional relations. We focus on programs that use discrete numerical distributions, but even then the space of possible outcomes may be uncountable (as a solution can be infinite). We define a probability measure over possible outcomes by applying the known concept of cylinder sets to a probabilistic chase procedure. We show that the resulting semantics is robust under different chases. We also identify conditions guaranteeing that all possible outcomes are finite (and then the probability space is discrete). We argue that the framework we propose retains the purely declarative nature of Datalog, and allows for natural specifications of statistical models.

In recent years, distributed representations of inputs have led to performance gains in many applications by allowing statistical information to be shared across inputs. However, the predicted outputs (labels, and more generally structures) are still treated as discrete objects even though outputs are often not discrete units of meaning. In this paper, we present a new formulation for structured prediction where we represent individual labels in a structure as dense vectors and allow semantically similar labels to share parameters. We extend this representation to larger structures by defining compositionality using tensor products to give a natural generalization of standard structured prediction approaches. We define a learning objective for jointly learning the model parameters and the label vectors and propose an alternating minimization algorithm for learning. We show that our formulation outperforms structural SVM baselines in two tasks: multiclass document classification and part-of-speech tagging.

We introduce a new approach to unsupervised estimation of feature-rich semantic role labeling models. Our model consists of two components: (1) an encoding component: a semantic role labeling model which predicts roles given a rich set of syntactic and lexical features; (2) a reconstruction component: a tensor factorization model which relies on roles to predict argument fillers. When the components are estimated jointly to minimize errors in argument reconstruction, the induced roles largely correspond to roles defined in annotated resources. Our method performs on par with most accurate role induction methods on English and German, even though, unlike these previous approaches, we do not incorporate any prior linguistic knowledge about the languages.