Artificial Intelligence Examples of the difficulty of NLP Levels of ...

9 downloads 50 Views 14KB Size Report
that success in NLP requires solving all of the most difficult problems of AI q Successful NLP requires a large common-sense knowledge base (CYC) q As with ...
Natural language processing (NLP) ●

Artificial Intelligence Today: An overview of some fascinating and important areas of AI that we have not had time to cover in this course Next class: A review of what we did cover in this course, in preparation for the final







CS 4633/6633 Artificial Intelligence

Examples of the difficulty of NLP ●

Machine translation: English Russian – “The spirit is willing but the flesh is weak” --> “The vodka is fine but the meat is rotten” – “Out of sight, out of mind” --> “The man is as blind as a saint”

CS 4633/6633 Artificial Intelligence

Levels of language analysis ●





● Sentence understanding depends on knowledge

about world – – – –

George: “I visited my aunt in the hospital and took her flowers.” Gracie: “That’s terrible. You should have brought her flowers!” “The pen is in the box.” “The box is in the pen.”





CS 4633/6633 Artificial Intelligence

● ●

Machine translation Natural language query languages for databases Text interpretation (more information is stored in text than in databases) – information retrieval: choose from a set of documents the ones relevant to a query – text categorization: e.g., categorize news stories by topic – information extraction from text CS 4633/6633 Artificial Intelligence

Morphology: Study of the structure of words, how words are formed from prefixes, suffixes, and other components Syntax: Study of the structure of sentences, how words combine into phrases, and phrases combine into sentences. A parser transforms a sequence of words into a structure that shows how the words are related, or rejects the sentence such as “Boy the go to the store” as ungrammatical Semantics: Study of meaning and how to assign meaning to the syntactic structures discovered by the parser. A grammatical sentence such as “Colorless green ideas sleep furiously” (Chomsky) may be rejected as semantically anomolous. Discourse: Study of the structure of extended (multiple sentence) language, such as in text or in dialogs. The meaning of an individual sentence may depend on the sentences that precede it and may influence the meaning of the sentences that follow it. Pragmatics: Study of the use of language, how sentences are used to convey information, make requests, and so on.

CS 4633/6633 Artificial Intelligence

Speech Recognition

Applications of NLP ●

Two aspects of NLP – natural language understanding – natural language generation NLP is sometimes said to be “AI-hard” in the sense that success in NLP requires solving all of the most difficult problems of AI Successful NLP requires a large common-sense knowledge base (CYC) As with expert systems, one way to make NLP practical is to focus on a specialized domain and particular task

● ●



Problem of inferring sentences from spoken sound This type of inference is highly uncertain and a probabilistic approach has performed best Hidden Markov models – rigorous mathematical theory base on probabilities – can be trained from a large corpus of speech data



Applications – automatic dictation – human/computer interaction CS 4633/6633 Artificial Intelligence

Computer Vision ●



● ●

Examples of vision tasks – navigation through a complex environment based on visual information (ALVINN) – recognition of classes of common objects (such as people or trees) in a complex scene An important problem is to derive an accurate 3-dimensional description of the objects in a scene and their properties, based on more or more 2-d images of the scene Some difficulties: perspective, shadow, occlusion, light source Understanding an image is similar to understanding language, and has similar difficulties (e.g., ambiguity) --> therefore, focus on specialized tasks CS 4633/6633 Artificial Intelligence

Levels of analysis Digitization and noise removal Edge detection, segmentation, and region finding ● Determine shapes and 3-d information ● (stereo vision and motion detection) ● Object and scene recognition ●



CS 4633/6633 Artificial Intelligence

Robotics

Applications of Computer Vision ●

Automated navigation – robotics – autonomous vehicles (ALVINN)

● ● ● ● ●

Visual inspection of integrated circuits Document reading Analysis of medical or satellite images Face recognition Human/computer interactions, e.g., recognizing the intentions and emotions of a computer user CS 4633/6633 Artificial Intelligence

Other areas of current research Smart rooms, ubiquitous computing ● Wearable computers ● Intelligent web assistants, “software agents” or “softbots” ● Intelligent tutoring systems ● Computers with emotions ●

CS 4633/6633 Artificial Intelligence

● ● ● ●

Complete agent: vision, planning (motion planning, navigation) Robot architectures: high-level planning (long-term deliberation) and low-level control (short-term reflexes) Lots of uncertainty Applications – manufacturing: replace humans in repetitive assembly-line work – replace humans in hazardous environments (handling toxic and radioactive materials) and remote environments (Mars, underseas exploration -- AUVs) – robot surgery: replace humans in tasks that require great precision – pets!

CS 4633/6633 Artificial Intelligence