{ "paper_id": "W83-0111", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T06:08:02.601690Z" }, "title": "", "authors": [ { "first": "Harri", "middle": [], "last": "Jsppinen", "suffix": "", "affiliation": { "laboratory": "", "institution": "Helsinki University of Technology Espoo", "location": { "country": "Finland" } }, "email": "" }, { "first": "Aarno", "middle": [], "last": "Lehtola", "suffix": "", "affiliation": { "laboratory": "", "institution": "Helsinki University of Technology Espoo", "location": { "country": "Finland" } }, "email": "" }, { "first": "Esa", "middle": [], "last": "Nelimarkka", "suffix": "", "affiliation": { "laboratory": "", "institution": "Helsinki University of Technology Espoo", "location": { "country": "Finland" } }, "email": "" }, { "first": "Matti", "middle": [], "last": "Ylilammi", "suffix": "", "affiliation": { "laboratory": "", "institution": "Helsinki University of Technology Espoo", "location": { "country": "Finland" } }, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "W83-0111", "_pdf_hash": "", "abstract": [], "body_text": [ { "text": "We are currently designing a data-base Interface for queries in natural Other models have been reported. Brodda and Karlsson (1980) attempted to find the most probable morphotactic segmentations for Finnish word forms without a reference to a lexicon. They report close to 90 $ accuracy.", "cite_spans": [ { "start": 105, "end": 131, "text": "Brodda and Karlsson (1980)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": null }, { "text": "System to a subclass of Finnish morphology. Karttunen et al. (1981) and Koskenniemi (1983) ", "cite_spans": [ { "start": 44, "end": 67, "text": "Karttunen et al. (1981)", "ref_id": null }, { "start": 72, "end": 90, "text": "Koskenniemi (1983)", "ref_id": "BIBREF6" } ], "ref_spans": [], "eq_spans": [], "section": "SSgvall-Hein (1978) reports an attempt to apply the Reversible Grammar", "sec_num": null }, { "text": "We approached the problem of morphological analysis from the vantage point of heuristic search. Heuristics has been used for many years in artificial intelligence problem solving situations. In more recent research heuristics has been expanded into multi-level search, and novel control strategies have been developed to govern the process. Examples are speech under standing (Erman et al., 1980 ) and many so called expert systems.", "cite_spans": [ { "start": 376, "end": 395, "text": "(Erman et al., 1980", "ref_id": "BIBREF1" } ], "ref_spans": [], "eq_spans": [], "section": "Heuristic morphology", "sec_num": null }, { "text": "Morphology is a much more constrained task domain than speech under standing, mass spectrometry, medical diagnosis, and other typical expert system applications. We, however, decided to study how multi-dimensional heuristic search applies to morphology. We do not use heuristic search because algorithmic methods would not apply (they do, as Karttunen et al. (1981) and Koskenniemi (1983) have demonstrated). Our argument for using heuristic rules was to see if we could get a faster method which would distribute most morphological knowledge in active rules.", "cite_spans": [ { "start": 342, "end": 365, "text": "Karttunen et al. (1981)", "ref_id": null }, { "start": 370, "end": 388, "text": "Koskenniemi (1983)", "ref_id": "BIBREF6" } ], "ref_spans": [], "eq_spans": [], "section": "Heuristic morphology", "sec_num": null }, { "text": "We knew that we must be on guard against risks involved: heuristics might sometimes erroneously generate wrong interpretations. Such dangers, however, did not materialize. ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Heuristic morphology", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "An experiment with automatic morphological analysis of Finnish, Papers from the Institute of Linguistics", "authors": [ { "first": "B", "middle": [], "last": "Brodda", "suffix": "" }, { "first": "F", "middle": [], "last": "Karlsson", "suffix": "" } ], "year": 1930, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Brodda, B., Karlsson, F., An experiment with automatic morphological analysis of Finnish, Papers from the Institute of Linguistics, University of Stockholm, Stockholm 1930.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "The Hearsay-II speech-understanding system; integrating knowledge to resolve uncertainty. Computing Surveys", "authors": [ { "first": "L", "middle": [ "D" ], "last": "Erman", "suffix": "" } ], "year": 1980, "venue": "", "volume": "", "issue": "", "pages": "213--253", "other_ids": {}, "num": null, "urls": [], "raw_text": "Erman, L.D. et al.. The Hearsay-II speech-understanding system; integrating knowledge to resolve uncertainty. Computing Surveys (June, 1980), 213-253.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Morphological Analysis of Finnish -A heuristic approach", "authors": [ { "first": "H", "middle": [], "last": "Jsppinen", "suffix": "" }, { "first": "A", "middle": [], "last": "Lehtola", "suffix": "" }, { "first": "E", "middle": [], "last": "Nelimarkka", "suffix": "" }, { "first": "M", "middle": [], "last": "Ylilarami", "suffix": "" } ], "year": 1983, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "JSppinen, H., Lehtola, A., Nelimarkka, E. and Ylilarami, M., Morphological Analysis of Finnish -A heuristic approach. Helslngi University of Technology, Digital Systems Laboratory, Report B 26, 1983.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "Morphological analysis of Finnish by computer", "authors": [ { "first": "L", "middle": [], "last": "Karttunen", "suffix": "" }, { "first": "R", "middle": [], "last": "Root", "suffix": "" }, { "first": "", "middle": [], "last": "Uszkoreit", "suffix": "" }, { "first": "", "middle": [], "last": "Hi", "suffix": "" }, { "first": "", "middle": [], "last": "Texfin", "suffix": "" } ], "year": null, "venue": "Proc. of the 71st Ann. Meeting of the SASS", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Karttunen, L., Root, R., and Uszkoreit, HI, TEXFIN; Morphological analysis of Finnish by computer. Proc. of the 71st Ann. Meeting of the SASS.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Two-level model for morphological analysis", "authors": [ { "first": "K", "middle": [], "last": "Koskenniemi", "suffix": "" } ], "year": 1983, "venue": "", "volume": "", "issue": "", "pages": "683--685", "other_ids": {}, "num": null, "urls": [], "raw_text": "Koskenniemi, K., Two-level model for morphological analysis. IJCAI-8 3 , 1983, 683-685.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Finnish morphological analysis in the reversible grammar system", "authors": [ { "first": "A-L", "middle": [], "last": "Sigvall-Hein", "suffix": "" } ], "year": 1978, "venue": "Proceedings of NODALIDA 1983", "volume": "78", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Sigvall-Hein, A-L., Finnish morphological analysis in the reversible grammar system, COLING 78, 1978. 120 120 Proceedings of NODALIDA 1983", "links": null } }, "ref_entries": { "FIGREF0": { "text": "As is well known, Finnish is a highly inflectional language. Con sequently, in data base applications as well as in other natural Finnish processing systems morphological analysis of word forms constitutes a fundamental computational subproblem. This paper describes our solution to the problem. Our model is intended for the analysis of Finnish word forms. The system performs all meaningful morphotactic segmentations for a given surface word form, transforms alternated stems into the basic form (sg nominative or 1st infinitive for nominals and verbs, respectively), and matches the stems against lexical entries in order to find the meaninful words. The present version of the system does not analyze compound word forma into their constituents, nor does Lt analyze derivational word forms. We are building a new version which will have some of these characteristics. Otherwise model is complete; it has been fully implemented, and tests indicate its correctness so far to lie in the neighborhood of 99.5 % (JSppinen et al., 1983).", "type_str": "figure", "num": null, "uris": null }, "FIGREF1": { "text": "report two distinct and complete models. Both systems first search in a lexicon all words whose roots match with a given input word form and then prune the ones which could result in the input word form. The latter model is symmetric: it analyzes as well as generates Finnish word forms. This research is supported by SITRA (Finnish National Fund for Research and Development), PL 329, 00121 Helsinki 12, Finland 111 Knowledge Engineering Applied to Morphological Analysis Harri J\u00e4 ppinen, Aarno Lehtola, Esa Nelimarkka, Matti Ylilammi Proceedings of NODALIDA 1983, pages 111-120 Due to the computational environment of our model, data-base interface, we set forth the following three design objectives for our analyser; 1) analysis should be efficient, 2) all valid interpretations of an input word form should be found (in the context of a given lexicon), and 3) the addition of new lexical entries should be easy. The last goal, a human engineering viewpoint, suggested us to minimize morphological information in lexical entries and, instead, store morphological data maximally into active knowledge sources.", "type_str": "figure", "num": null, "uris": null }, "FIGREF2": { "text": "+ $ + number + case + (p o s s es sio n ) + ( c l i t i c s ) . Root denotes the unvarying head part of a word stem, and the stem ending ($) its alternating tail. The root may, however, vary under consonant gra dation process. Finnish nominals appear in 1U cases: nominative, genetive, partitive, essive, translative, instructive, abessive, ablative, and allative. Genetive, partitive and illative are more irregular than the others in that they realize in more than one allomorphs. Plural is indicated by a 'i ', 'j', 't' or a null string (^) depending on a context. To visualize, the Finnish word forms 'takissaniko* (= in my coat?) and \u2022takkeihisihan' (= into your coats!) are segmented as (forms may receive some but not all cases, and the 1st participle may participate in the adjective comparation process. Most nomi nal forms may receive a person and a clitic segment in the standard way. A comparative adjective in Finnish is indicated by the suffix 'structure of MORFIN, as we call the model, employs the hypothesis-and-test paradignas follows. A global data base is divided into four levels; surface word form (SWF), morphotactic (MT), basic word form (BWT) and confirmation (C) levels. Between these levels active knowledge sources, production rules, progressively generate and test hypotheses. Between the surface word form level and the morphotactic level morphotactic productions (MPs) produce hypotheses of possible segmentations and interpretations of an input word. They leave alternated stems untouched. Stem productions (SPs) produce inverse transformations of the variant stems into canonical basic word forms. For nominals we use the singular nomina tive case and for verbs the 1st infinitive as the basic form. Some stems get rejected in this second phase as impossible alternations. Dictionary look-up finally tests the proliferated hypothetical basic word forms against the existing lexicon. Morphologically ambiguous words result in multiple confirmation level entries, if the words exist in the lexicon.", "type_str": "figure", "num": null, "uris": null }, "FIGREF3": { "text": "portrays the levels and an example analysis of an ambiguous sur face word form. The numbers in parentheses are pointers to the previous level. *\u2022' indicates a confirmed hypothesis; 'VA' and 'HA' are postulates for the strong (or neutral) and weak (or neutral) grades, respectively. MORFIN is embedded in the MPs. The morpheme pro ductions recognize legal morphological surface-segment configurations in a word and slice the word accordingly. The recognition proceeds from right to left up to the stem boundary as in Brodda and Karlsson (1981). That is, for nominals, verbs and adjectival comparatlon forms morphotactic segmentations are done from clitics up to (but excluding) the stem ending. We use directly the allomorphlc variants of the morphemes. Since possible segment configurations overlap, several mutually exclusive hypotheses are usually produced on the morphotactic level. We dressed the morphotactical knowledge of MORFIN into context-sensitive rules. The condition part in a production recognizes a single valid morpho tactic segment. To prune search we attached up to two left currently 201 distinct MPs. To facilitate efficient processing we compiled the MPs into 32 distinct state transition automata (3 for clitics, 1 for person, 5 for tense, 3 for case, postulate canonical basic forms which in the context of the given morphotactic segmentations might have resulted in the observed variant stem forms. The rules may reject a candidate stem form as an impossible transformation, or produce one or more basic form hypotheses. true if 4-| and &2 and ... and 4^ are true, in that order, under the stipulation that the recognized letters in a stem are consumed (the next condition tests the next letter).", "type_str": "figure", "num": null, "uris": null }, "FIGREF4": { "text": "production recognizes, among others, the genetive stems 'kahde' (=of two) or 'yhde* (=of one) and generates basic word form hypotheses *ka ksi' (=two) and 'y ksi' (=one), respectively. The SPs were collected into 11 distinct sets of productions for nominals and 6 sets for verbs. On average a set has about 25 rules. These sets were compiled into state transition automata to yield efficient processing. . An entry matches with a hypothesis if the words match and the grades are not of the opposite strength. If the hypothesis is an adjective comparation form, the lexical entry must furthermore be marked as an adjective. whole Finnish morphology with the provision below. The singular instructive and the old plural genetive IN-cases are exceedingly rare and are left out for the sake of efficiency. They could be easily added. Currently compound nouns are not analyzed into their constituent parts but we are in the process of designing a new version of MORFIN which will analyze also compound nouns. There is a precompiled version of MORFIN written in standard Pascal. It has been tested both in DEC 20 and VAX 11/780 configurations. The analysis of a random word in a newspaper text takes between U to 60 ms of DEC 2060 CPU time, about 15 ms on average. On average about 4 postulates were generated on the basic word form level. The basic approach of MORFIN applies well, we believe, also to analysis of derivational word forms. All one has to do is to add proper MPs and (sometimes) SPs (if needed). Thus, for instance, to account 'iloinen' -> 'iloisesti' derivation we have to add only a single MP to recognize *sti* and can use the already implemented SPs: the stem alternation is similar to the singular elative case: 'iloinen* (=glad) -> 'iloisesta' (=from glad). One of the main objectives in our design was to store minimal amount of morphological data in a lexicon and dress it maximally in active rules. We feel success in this regard. The only morphological knowledge words in a lexicon carry is the boundary between the root and the stem ending of a word and the grade of the stem, entries. Consequently, MORFIN has a con venient functional feature: words not existing in a lexicon get analyzed as well as those that do exist. A user can add new entries simply by indi cating which of the postulated forms are right. It seems that the introduc tion of new lexical entries is not as straightforward for a casual user in the other systems. It seems implausible to us that a native speaker of an inflectional language tags morphological data in individual words. If we take a gram matical but meaningless (non-existent) word, say, 'ventukoissa* and test native Finns, they probably all would agree that it represents the plural inessive form of a non-existent word 'ventukka'. Our model covers such phe nomena. 'Ventukoissa' is analyzed as well as meaningful words. Only the dictionary look-up process rejects the word as meaningless. and implemented a system for the computational morphologi cal analysis of Finnish word forms. Our analysis is based on multi-level heuristic search in which modular active knowledge sources postulate and evaluate partial morphological interpretations. On the first level raorphotactic productions postulate and interprete raorphotactic segmentations of an input word. The second phase converts the postulated variant stems into their basic forms. The third phase matches the proliferated basic word form hypotheses against a lexicon. All morphological knowledge other than the root boundary and the grade of a lexeme is dressed in procedural form, which yields efficient analysis. Grammatical but meaningless word forms become analyzed in the model as well as meaningful ones.", "type_str": "figure", "num": null, "uris": null } } } }