Performance Evaluation for Text Processing of Noisy Inputs Daniel - PowerPoint PPT Presentation

Performance Evaluation for Text Processing of Noisy Inputs Daniel Lopresti Computer Science & Engineering Lehigh University Bethlehem, PA 18015, USA Performance Evaluation for Text Processing of Noisy Inputs Daniel Lopresti

Motivation Earlier attempt to study impact of errors from optical character recognition (OCR) on automatic summarization:  use shallow language understanding (tokenization, PoS tagging),  apply past statistics (word frequencies, sentence positions),  look for cue phrases (e.g., “In conclusion ...”),  extract key sentences (or phrases or paragraphs) for summary. Situation might arise when processing large quantities of scanned documents (e.g., information extraction, digital libraries). But many of same basic text processing steps apply elsewhere. “Summarizing Noisy Documents,” H. Jing, D. Lopresti, and C. Shih, Cut-and-paste Text Summarization , H. Jing, Ph.D. Thesis, Proceedings of the Symposium on Document Image Understanding Dept. of Computer Science, Columbia University, 2001. Technology , April 2003, Greenbelt, MD, pp. 111-119. Performance Evaluation for Text Processing of Noisy Inputs Daniel Lopresti

On Clean Input ... Ideal Summary (Human) Ideal Summary (Human) Kingdom To Sign Nuclear Non-proliferation Treaty Kingdom To Sign Nuclear Non-proliferation Treaty Kingdom To Sign Nuclear Non-proliferation Treaty Saudi Arabia on Tuesday decided to sign the nuclear weapons Saudi Arabia on Tuesday decided to sign the nuclear weapons Saudi Arabia on Tuesday decided to sign the nuclear weapons non-proliferation treaty, a strong indication it will not seek nuclear non-proliferation treaty, a strong indication it will not seek nuclear non-proliferation treaty, a strong indication it will not seek nuclear Saudi Arabia on Tuesday decided to sign the nuclear weapons Saudi Arabia on Tuesday decided to sign the nuclear weapons warheads for intermediate-range missiles it recently acquired warheads for intermediate-range missiles it recently acquired warheads for intermediate-range missiles it recently acquired non-proliferation treaty, a strong indication it will not seek nuclear non-proliferation treaty, a strong indication it will not seek nuclear from China. from China. from China. warheads for intermediate-range missiles it recently acquired warheads for intermediate-range missiles it recently acquired The official Saudi Press Agency reported that King Fahd made the The official Saudi Press Agency reported that King Fahd made the The official Saudi Press Agency reported that King Fahd made the from China. from China. decision during a Cabinet meeting in Riyadh, the Saudi capital. decision during a Cabinet meeting in Riyadh, the Saudi capital. decision during a Cabinet meeting in Riyadh, the Saudi capital. It appeared the timing of the decision was designed primarily to It appeared the timing of the decision was designed primarily to The meeting was called in response to a recommendation by The meeting was called in response to a recommendation by The meeting was called in response to a recommendation by reassure the United States that the kingdom will not try to arm its reassure the United States that the kingdom will not try to arm its Prince Saud al-Faisal, the Saudi foreign minister, that the Prince Saud al-Faisal, the Saudi foreign minister, that the Prince Saud al-Faisal, the Saudi foreign minister, that the CSS-2 missiles with nuclear warheads. CSS-2 missiles with nuclear warheads. kingdom sign the international treaty against the spread of kingdom sign the international treaty against the spread of kingdom sign the international treaty against the spread of nuclear arms. nuclear arms. nuclear arms. An account of the Cabinet discussions and decisions at the An account of the Cabinet discussions and decisions at the An account of the Cabinet discussions and decisions at the meeting, which ended before dawn, was issued by Information meeting, which ended before dawn, was issued by Information meeting, which ended before dawn, was issued by Information Minister Ali al-Shaer and distributed by the agency. The agency, Minister Ali al-Shaer and distributed by the agency. The agency, Minister Ali al-Shaer and distributed by the agency. The agency, Automatic Summary monitored in Bahrain, did not elaborate. monitored in Bahrain, did not elaborate. monitored in Bahrain, did not elaborate. It appeared the timing of the decision was designed primarily to It appeared the timing of the decision was designed primarily to It appeared the timing of the decision was designed primarily to Saudi Arabia on Tuesday decided to sign the nuclear weapons reassure the United States that the kingdom will not try to arm its reassure the United States that the kingdom will not try to arm its reassure the United States that the kingdom will not try to arm its non-proliferation treaty, a strong indication it will not seek nuclear CSS-2 missiles with nuclear warheads. The decision also was CSS-2 missiles with nuclear warheads. The decision also was CSS-2 missiles with nuclear warheads. The decision also was warheads for intermediate-range missiles it recently acquired viewed as an attempt to blunt Israel's allegations that the viewed as an attempt to blunt Israel's allegations that the viewed as an attempt to blunt Israel's allegations that the from China. missiles constituted a threat to its safety. missiles constituted a threat to its safety. missiles constituted a threat to its safety. Saudi Arabia, the Middle East petroleum giant and the world's Saudi Arabia, the Middle East petroleum giant and the world's Saudi Arabia, the Middle East petroleum giant and the world's Saudi Arabia, the Middle East petroleum giant and the world's largest exporter of crude oil, was reported to have recently largest exporter of crude oil, was reported to have recently largest exporter of crude oil, was reported to have recently largest exporter of crude oil, was reported to have recently acquired from Beijing an undisclosed number of CSS-2 missiles acquired from Beijing an undisclosed number of CSS-2 missiles acquired from Beijing an undisclosed number of CSS-2 missiles acquired from Beijing an undisclosed number of CSS-2 missiles capable of reaching virtually any point in the Middle East, capable of reaching virtually any point in the Middle East, capable of reaching virtually any point in the Middle East, capable of reaching virtually any point in the Middle East, including Israel. including Israel. including Israel. including Israel. Israel has voiced fears the Saudis might be seeking to acquire Israel has voiced fears the Saudis might be seeking to acquire Israel has voiced fears the Saudis might be seeking to acquire nuclear warheads for the missiles and indicated it might deal a nuclear warheads for the missiles and indicated it might deal a nuclear warheads for the missiles and indicated it might deal a preemptive blow. preemptive blow. preemptive blow. Performance Evaluation for Text Processing of Noisy Inputs Daniel Lopresti