Notes on the beauty of Mathematics: Jarinik and Modern language processing

In previous notes, especially in the statistical language model, we often mentioned Jarinik's name, and Dr. Friedrich Wilderness Jarinik was the master of the successful application of mathematical principles to the field of natural language processing.

Dr. Jarinik was born on November 18, 1932 and died on September 14, 2010. Born in the Czech Republic, he went through World War II, his father was a doctor, died in a concentration camp during the war, and in 1946 his mother immigrated to the United States with her family.

Jarinik's first dream was to become a lawyer, defend his father, immigrate to the United States, hoping to become a doctor, he wanted to go to Harvard Medical School, but he could not afford 8 years of tuition (4 years plus 4 years of medical education). And just as MIT offered him a full scholarship (for Eastern European immigrants), Jarinik chose MIT to study electrical engineering. This opens up his first step towards success.

At MIT, he met many world-class masters, including the originator of information, Dr. Shannon, a master of Linguistics, Jacobson (the famous six elements of communication). Later, Jarinik's wife came to the United States from the Czech Republic, at Harvard University, Jarinik often attend Harvard. There he often listens to the lessons of the great linguist Chomsky. The three masters have a great influence on Jarinik's later research direction-the use of information theory to solve language problems.

    Every person who makes speech recognition and language processing is familiar with the phrase: every day I expel a linguist, the recognition rate of my speech recognition system increases a bit. It comes from a story of Jarinik, who has been teaching at Harvard for a year after he received his doctorate from MIT, and then taught at Cornell University as Professor Jarinik. He chose Cornell because he had a good rapport with a linguist, Hackett, when he was looking for a job. At that time the professor expressed his readiness to cooperate with Jarinik in the use of the information theory ending pronunciation. But then the professor broke his promise. Jarinik's bad impression of the linguist began. Later in IBM found that linguists are well-organized, but the work is not high and low, to the linguists from this abhorrence, and then have the phrase " the famous .

In 1972, Jarinik to the IBM Watson Lab and inadvertently led the speech recognition lab, and switched to IBM two years later. where he formed a strong team of researchers, including his famous partner Bohr, the founder of the famous speech recognition Dragon, Baker and his wife, the two other co-authors of the BCJR algorithm for solving the maximum entropy iterative algorithm, cook and Tel Aviv, And the first Brown to propose a statistical model of machine translation. Even the most junior our crew task Rafat is now a well-known scholar.

In IBM, Jarinik and others put forward the framework structure of statistical speech recognition. Before Jarinik, the scientists used speech recognition as artificial only to match the model with the problem. and Jarinik it as a communication problem, and uses two hidden Markov models (acoustic model and language model) to generalize the speech recognition. This framework has far-reaching effects on speech and language processing so far, which not only makes it possible to use speech recognition fundamentally, but also lays the foundation for today's natural language processing.

One of the great contributions of Jarinik and Bohr, Cook and Tel Aviv to mankind is the BCJR algorithm, which is one of the two most widely used algorithms in mathematical communication today (the other is the Viterbi algorithm). However, 20 years after the invention of the algorithm, it was widely used. IBM has listed it as one of IBM's greatest contributions to humanity in history and is affixed to the wall of the California Amoton Laboratory. However, later BCJR's four people all left IBM.

Jarinik and IBM, a group of the most prominent scientists left IBM in the early 90, most of them later became tens of millions or even billionaires, but Jarinik is not a man to idle, 1994, he went to John·Hopkins University has established the world-renowned CLSP (Center for Language and speed processing) laboratory. In Jarinik toJohn·before Hopkins, this is why medical universities are not top notch in engineering, and there are no emerging disciplines such as speech recognition and natural language processing. Jarinik starts from scratch,In just two or three years, CLSP has become the world's leading research center.

Jarinik is still sharp after the age of more than 70 and goes to work on time every day. On September 14, 2010, he died of a heart attack at his desk. He began to create the largest speech and language processing center in the world's academia, and worked on the last day of his life, when others retired and spent their old age. Because he has a large number of students and friends working at Google, these people and Google for John Hopkins donated a sum of money and founded the Jarinik Scholarship.

Notes on the beauty of Mathematics: Jarinik and Modern language processing