Think a language-learning robot sounds like science fiction? The day may not be as far off as it seems, in light of new software, developed by Prof. Shimon Edelman, psychology, with colleagues from Tel Aviv University in Israel.
The soon-to-be-patented program – “Automatic Distillation of Structure,” or “ADIOS,” for short – can derive a language’s rules of grammar, and then produce sentences of its own, simply from blocks of text in that language.
“When scanning new input, the program looks for recurring patterns or interchangeable sequences,” explained Edelman, currently on sabbatical. “For example, if the following three sentences appeared in a [text] – ‘I saw a film today, oh boy,’ ‘He saw a film today at the reception,’ ‘She saw a film today and liked it,’ – the program would identify the sequence, ‘saw a film today,’ and determine whether it’s a statistically significant pattern. If so, the sequence is added to the software’s lexicon and can be used to create new sentences.”
The Israeli-born Edelman and his Tel Aviv University partners, Professors David Horn and Eytan Ruppin, and doctoral student Zach Solan, heralded the success of ADIOS in a paper, “Unsupervised Learning of Natural Languages,” published recently in the Proceedings of the National Academy of Sciences (PNAS, Vol. 102, No. 33).
Because the program’s algorithm relies on overlapping sequences in different sentences, Edelman says it has been shown to work best on languages with rigid syntax, like English and German. He claims the software can be refined to address shortcomings on more problematic languages, like his native Hebrew.
ADIOS is not limited to human language, however. The program has also detected patterns in sequences of non-linguistic data, such as musical scores and DNA strands.
Edelman believes this breakthrough will have a number of practical applications, most immediately for speech-recognition software. He says that ADIOS will improve upon existing units, like Amtrak’s virtual travel agent “Julie,” by giving them the grammatical knowledge to “guess what’s coming next in a sentence” and “learn how to overcome various obstacles to understanding, such as foreign accents.”
Edelman also thinks his findings may force linguists “to revise some of their preconceptions regarding language acquisition by children, language competence in adults, and second-language instruction.”
“We’ve modeled a couple of experiments from developmental psycholinguistics. People in those fields would be well-advised to look at our results,” he said.
“Apart from testing our algorithm extensively on standard benchmarks used in computational linguistics, we have demonstrated its ability to learn from realistic children-directed language (the CHILDES corpus). In addition, we have modeled several psycholinguistic experiments in artificial language acquisition. Our results suggest that the currently prevalent theories concerning linguistic development, language competence and processing, and second-language instruction will have to be revised.
The development of ADIOS was financed in part by the U.S.-Israel Binational Science Foundation, established in 1972 by the two countries’ governments. Technological cooperation between American and Israeli companies and researchers has become increasingly common in recent years.
Products of that collaboration include the cell phone, Intel’s Pentium IV and Centrino processors, Windows XP and AOL Instant Messenger.
Cornell itself recently partnered with the State of Israel and the Hashemite Kingdom of Jordan to establish the Bridging the Rift Center, a scientific research facility, to be built on the Israeli-Jordanian border.
Some campus leaders believe the development of ADIOS illustrates the value of Cornell’s continued participation in American-Israeli technological enterprise: “This scientific breakthrough shows us once again the benefits of investment and cooperation with Israel,” said Student Assembly President Tim Lim. “More joint endeavors, such as ADIOS and the Bridging the Rift Center, should be encouraged, and I hope Cornell University will be at the forefront of this American-Israeli technological cooperation.”
Archived article by Ben Birnbaum
Sun Staff Writer