| { |
| "paper_id": "C86-1033", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T13:15:22.806137Z" |
| }, |
| "title": "A Stochastic Approach t O Parsing", |
| "authors": [ |
| { |
| "first": "Geoffrey", |
| "middle": [], |
| "last": "Sampson", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Leeds", |
| "location": {} |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "", |
| "pdf_parse": { |
| "paper_id": "C86-1033", |
| "_pdf_hash": "", |
| "abstract": [], |
| "body_text": [ |
| { |
| "text": "Simulated annealing (e.g. Kirkpatrick et alo 1983 , Bridle & Moore 1984 , Ackley et al. 1985 is a stochastic computational technique for finding optimal solutions to combinatorial problems for which the combinatorial explosion phenomenon rules out the possibility of systematically examining each alternative.", |
| "cite_spans": [ |
| { |
| "start": 26, |
| "end": 49, |
| "text": "Kirkpatrick et alo 1983", |
| "ref_id": null |
| }, |
| { |
| "start": 50, |
| "end": 71, |
| "text": ", Bridle & Moore 1984", |
| "ref_id": null |
| }, |
| { |
| "start": 72, |
| "end": 92, |
| "text": ", Ackley et al. 1985", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I.", |
| "sec_num": null |
| }, |
| { |
| "text": "It is currently being applied to the practical problem of optimizing the physical design of computer circuitry, and to the theoretical problems of resolving patterns of auditory and visual stimulation into meaningful arrangements of phonemes and three-dimensional objects. Grammatical parsing --resolving unanalysed linear sequences of words into meaningful grammatical structures --can be regarded as a perception problem logically analogous to those just cited, and simulated annealing holds great promise as a parsing technique.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "I.", |
| "sec_num": null |
| }, |
| { |
| "text": "annealing can most directly be explained via a physical analogy.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Simulated", |
| "sec_num": null |
| }, |
| { |
| "text": "Consider the logical space of alternatiw~s in some large-scale combinatorial problem as a chunk of mountainous terrain, in which the altitude of any point corresponds to the relative \"goodness\" of that particular solution (the lower, the better).", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 9, |
| "end": 124, |
| "text": "the logical space of alternatiw~s in some large-scale combinatorial problem as a chunk of mountainous", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Simulated", |
| "sec_num": null |
| }, |
| { |
| "text": "We want to find the lowest point, but there are far too many points for each to be considered separately.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Simulated", |
| "sec_num": null |
| }, |
| { |
| "text": "We might try to locate the low point by dropping a ball onto the territory at random and hoping it will roll down to the low point. This corresponds to randomly choosing a particular overall solution to the combinatorial problem, and then considering a series of modifications to individual components of the solution, adopting the modifications whenever they improve the overall solution and rejecting them otherwise. But the ball is very unlikely to reach the low point.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Simulated", |
| "sec_num": null |
| }, |
| { |
| "text": "Much more probably, it will roll a short way downhill and qome to rest in a \"local minimum\", a place .where all immediate moves are uphill even though, some distance away, there are places much lower than the spot where the ball has halted. This is simulated annealing.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Simulated", |
| "sec_num": null |
| }, |
| { |
| "text": "Not all combinatorial problems are amenable to the annealing technique.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Simulated", |
| "sec_num": null |
| }, |
| { |
| "text": "If the desired low point in the mountainous terrain of the analogy happened to be at the bottom of a deep, narrow mineshaft sunk from a high place, annealing would not help to find it. But the logical geometry of many real-life combinatorial phenomena is more like the geometry of natural mountains, where there is a strong tendency for relatively low-lying points to be adjacent to many other relatively low-lying points.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Simulated", |
| "sec_num": null |
| }, |
| { |
| "text": "For such phenomena, simulated annealing can be an efficient way of arriving at optimal solutions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Simulated", |
| "sec_num": null |
| }, |
| { |
| "text": "The applicability of annealing as a parsing technique presupposes a statistical approach to NL analysis which will itself be unfamiliar to many readers. At this point I must therefore digress from the annealing concept in order briefly to describe the statistics-based NL research paradigm within which I am working, and to which simulated annealing appears to offer an important contribution. The initial temperature (standard deviation of the Gaussian) was I.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "2.", |
| "sec_num": null |
| }, |
| { |
| "text": "The temperature was reduced by 3% after every fiftieth successive attempt to change the tree~ The system was deemed to have \"frozen\" at the first temperature-drop at which each of the 100 preceding attempts to change the tree either had been rejected or left the value of the tree unchanged. Yet the range of possibilities out of which the system has winnowed the correct analysis includes massive numbers of utterly crazy structures: note for instance how in the early stages of the run illustrated the system has considered a tree including a genitive phrase (G) consisting of a finite clause (F) followed by an adjective phrase (J) --a constituent which linguistically makes no sense at all.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "2.", |
| "sec_num": null |
| }, |
| { |
| "text": "Considering how many alternative logically-possible solutions are available to the system, a few thousand steps seems a small number by which to reach the correct solution or even its vicinity. In a dynamic annealing parser, all parts of the curreDtr tree would at all stages be available to revision, but the relative rarity of the need for backtracking could be exploited by adding a bias to the function which randomly selects nodes for reconsideration, so that nodes are reconsidered less frequently as they become \"older\".", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "2.", |
| "sec_num": null |
| }, |
| { |
| "text": "Since the bulk of computing time in an annealing parser would undoubtedly be consumed in calculating gains and losses for candidate tree-changes, this system of concentrating the search for profitable tree-changes on the areas of the tree where such changes are most likely to be found could be a good means of saving processing time by reducing the total number of moves considered.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "2.", |
| "sec_num": null |
| }, |
| { |
| "text": "A ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "7.", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": {}, |
| "ref_entries": {} |
| } |
| } |
