| { |
| "paper_id": "W83-0115", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T06:07:59.307482Z" |
| }, |
| "title": "Gunnel. K\u00e4llgren Institutionen f\u00f6r lingvistik. Stockholms universitet 106 91 Stockholm HP -A HEURISTIC FINITE STATE PARSER BASED ON MORPHOLOGY", |
| "authors": [ |
| { |
| "first": "In", |
| "middle": [], |
| "last": "Che", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "", |
| "pdf_parse": { |
| "paper_id": "W83-0115", |
| "_pdf_hash": "", |
| "abstract": [], |
| "body_text": [ |
| { |
| "text": "To he able to explain the ability of human beings to perceive and interpret language as quickly and as correctly as Is actually done, we have to make certain assumptions. We must e g assume that humans have a highly developed skill for parallel pro Classing, not only in the sense that one can take part in a conversation while driving a car and rubbing one's nose at the same time-More important In this connection Is that language perception and language understanding can be fruitfully regarded as a system of parallel processes.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Is that most of these processes work In a heuristic manner, in the sense that they strive for optimal probability rather than absolute correctness, that in choice situations they take a fair guess or a short-cut rather than a toilsome testing of every alternative.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Another Important assumption", |
| "sec_num": null |
| }, |
| { |
| "text": "our HP-project, we try to draw some consequences of these assumptions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "In", |
| "sec_num": null |
| }, |
| { |
| "text": "For a start, we have picked out what we believe to be one of the many processes Involved and furthermore we treat it as consisting In Its turn of several parallel subprocesses.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "In", |
| "sec_num": null |
| }, |
| { |
| "text": "We also strive to make it work \"heur1st1ca1ly\" In the way sketched above, 1 e more by trial and error than by careful stepwise analysis. In this way we will try to build and study a model that will perform a few of the many tasks a human mind can manage.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "In", |
| "sec_num": null |
| }, |
| { |
| "text": "major process we have chosen has to do with perception and interpretation of surface morphological features. This, we think, plays an Important role In word class assessment and deciding of constituent structure. To be able to really Interpret the full text, enormous amounts of semantic as well as pragmatic information and general knowledge of the world Is necessary.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The", |
| "sec_num": null |
| }, |
| { |
| "text": "This, we do not even try to do. Instead we want to stick strictly to a purely morphological surface level in order to Investigate how much information about language structure con be gathered from that level. It may not be possible to say precisely HOW much, but we have already found that it Is surprisingly much, so much that it has to be payed more heed to it also In more lexically based models.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The", |
| "sec_num": null |
| }, |
| { |
| "text": "To give a technical overview of the system, it can be described as a series of Finite State Pattern Matchinv; machines (FS-PM), where the superordinate system governing the subprocesses can also be regarded as a Finite State machine. This will make the system equivalent to an Augmented Transitory Network. text can be taken and processed in a short time and to a low cost, but the results will improve somewhat if some rules are designed to handle the peculiarities of a specific text type, e g rules for legal texts as opposed to fictional prose. \"provides\". Words with a prefix and ending in a vowel are too ambiguous to be marked yet, like F0R>KLAR'A infinite verb \"explain\", FOR>m AG'A indefinite noun \"ability\", FOR^SYNT A adjective \"timid\". Words with a recognizable adjective ending were identified already in SM, F0R>KLAR<L1G'A \"explicable , as were preterite verbs, FOR>KLAR=ADE \"explained\". All other words with a prefix are simply assumed to be nouns. This holds to a remarkably high degree but is a point where we have to swallow some errors in the hope that they will be amended further on.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The", |
| "sec_num": null |
| }, |
| { |
| "text": "also contains some other types of possible word stems that are treated in a similar way. Some outputs from PF: ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "PF", |
| "sec_num": null |
| }, |
| { |
| "text": "Proceedings of NODALIDA 1983", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [ |
| { |
| "text": "F0R>STA \"understand\". This Is again based on probabilities, but the pattern \"possible trigger (+ allowed constituent) + word of the described form\" Is strong enough to pick out almost every Infinite verb with very few overgenerat1ons. Chains of auxiliary verbs can also be managed. Illustrations: A word like f AnG=AR will have been marked with 'w'by SM, as it can be either a verb in present tense, \"catches\", or a plural noun, \"prisoners\". If nothing else has changed the marking during the processing. It is likely that DA will. 10w FANG=ARw v FLYDDEv . 11aDARlFRANa vFLYDDEv w FAn g =ARw (12) nFLICK=ANn w FAn g =ARw nHUND=ENn prisoners escaped' \u25a0from there escaped prisoner the girl catches the dog' In (10) and (11) 'w' appears before or after 'v', e g a finite verb. Two finite verbs are not allowed In the same clause and there Is nothing to signal clause boundary, so the w-word must be a noun. In (12) 'w' both precedes and follows a noun. Either condition would be sufficient to turn 'w' Into 'v' for finite verb, as noun phrases are very rarely given on a row with nothing between them. DA contains several similar mechanisms for changing earlier marks. In case of discrepancies, the semantics will probably always win, but where the results are in accordance, which they mostly are, the morphological Interpreter may well be the process that gives sentences their grammatical structure.If we regard the HP-systera as a first attempt at modelling this morphological Interpreter it means that the system also has a psycho1ingu1stic relevance that deserves to be seriously considered and investigated.K\u00e4llgren, Gunnel (1983): Substantivjakten.To be published as IRI-PM, Institutet f\u00f6r R\u00e411 s 1 nf orma 11k , Stockholms universitet.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "annex", |
| "sec_num": null |
| } |
| ], |
| "bib_entries": {}, |
| "ref_entries": { |
| "FIGREF1": { |
| "uris": null, |
| "type_str": "figure", |
| "text": "follow covers the state of the system in tVu; autumn of 1983. It is important to remember that the system is still only in the first part of its development. There Is much left to of lexicon rules in the system is at present about 300. Most of them are in the first program, MK (Mark). MK marks the words for word class according to the code given code letter is put before and after the identified word, thus making up bracket and mark in one. derivational and flexlonal suffixes are segmented and marked, as are some segmentable final letters. If information about word class can be extracted from the Internal structure of a word, \"bracket\" marking according to the code in Figure 1 is also entered. Words ending in e g NING or SION/TION (plus possible flexlonal endings) are always nouns in Swedish and get marked as such by Words with a prefix and ending in vowel + R are also supposed to be verbs, this time in the present tense, as F0R>SER", |
| "num": null |
| }, |
| "FIGREF2": { |
| "uris": null, |
| "type_str": "figure", |
| "text": "high combinab111ty. Even so, the intervening adjectives must always be checked EN GLAD BAG=ARE \"a happy baker\" is OK, as is N\u00c4GRA segmented prefix and a monosyllabic stem as above. 159 159 Proceedings of NODALIDA 1983", |
| "num": null |
| }, |
| "TABREF0": { |
| "html": null, |
| "text": "", |
| "content": "<table><tr><td>Each</td><td>of</td><td>the</td><td colspan=\"3\">subordinate</td><td colspan=\"2\">FS-PM</td><td colspan=\"2\">machines</td><td>corresponds</td><td>to</td><td>a</td></tr><tr><td colspan=\"2\">separate</td><td colspan=\"2\">computer</td><td colspan=\"2\">program.</td><td colspan=\"2\">For</td><td colspan=\"2\">computational</td><td>reasons</td><td>the</td></tr><tr><td colspan=\"2\">programs</td><td colspan=\"8\">are run input</td><td>and</td></tr><tr><td>output</td><td/><td>are</td><td>run,</td><td>but</td><td colspan=\"2\">apart</td><td>from</td><td/><td>that, no pre-processing is</td></tr><tr><td colspan=\"7\">needed. A full lexicon is not</td><td colspan=\"2\">needed</td><td>either.</td><td>Literally</td><td>any</td></tr><tr><td colspan=\"2\">Swedish</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td/><td/><td/><td/><td/><td>155</td><td/><td/></tr><tr><td colspan=\"10\">HP -A Heuristic Finite State Parser Based on Morphology</td></tr><tr><td>Gunnel K\u00e4 llgren</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"7\">Proceedings of NODALIDA 1983, pages 155-161</td><td/><td/></tr></table>", |
| "num": null, |
| "type_str": "table" |
| } |
| } |
| } |
| } |
