File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/87/j87-3003_metho.xml
Size: 110,501 bytes
Last Modified: 2025-10-06 14:11:58
<?xml version="1.0" standalone="yes"?> <Paper uid="J87-3003"> <Title>TOOLS AND METHODS FOR COMPUTATIONAL LEXICOLOGY</Title> <Section position="3" start_page="0" end_page="0" type="metho"> <SectionTitle> 2. ACQUISITION AND STORAGE OF MACHINE-READABLE DICTIONARIES. 2.1, NORMALIZING MACHINE-READABLE DICTIONARIES. </SectionTitle> <Paragraph position="0"> Once machine readable dictionary tapes are acquired from the publishers, they usually have to be cleaned up and/or normalized. We have encountered three basic types of publisher tapes: one type, represented by Longman and Webster's Seventh, has some information segmented into separate records labelled with an identifying code and other information implicit in font change codes. A second type, represented by Collins, consists of unsegmented text interspersed with font-change codes. A third type will be the new computerized OED (Tompa (1985)), where each type of information is specifically delimited at the beginning and end, and where there are any number of nesting levels. Much of the work described here was done with data of the second type; however, it is applicable also to the first type, and of course to the third.</Paragraph> <Paragraph position="1"> In the simplest sense, cleaning up usually means the removal of typesetting codes and other formatting information, or possibly removal of errors (usually done by hand). However, since the typesetting codes (e.g.</Paragraph> <Paragraph position="2"> begin-italic, begin-roman) are clues to the kind of information that follows the codes (for example, part-of-speech information appears in italics in the Collins dictionaries), they cannot simply be washed out. Instead, the raw data undergoes a process of normalization, the goal of which is to identify and label all of the information in the dictionary tapes in such a way that it can be easily accessed without detailed knowledge of the formatting conventions. This is our rationale for the creation of lexical data bases.</Paragraph> <Paragraph position="3"> We have carried out this normalization process in two stages. The goal of the first stage is to segment the information on the tapes into separate entries, storing the results in a direct access file, where the key is the dictionary headword and the data is simply the body of the entry. Dictionary data is stored in a Dictionary Access Method (DAM) file, from which it can be conveniently retrieved, by headword, for subsequent processing. DAM is completely described in Byrd, et al.(1986b). For this discussion, we note that DAM allows the efficient storage and retrieval of entries for words in alphabetical order and in random order by headword. Having the dictionary in a DAM file allows its entries to be accessed through the WordSmith on-line dictionary system, to be described later.</Paragraph> <Paragraph position="4"> The goal of the second stage of the normalization process is to segment and identify the data in the entry, creating a second DAM file in lexical data base format.</Paragraph> <Paragraph position="5"> Both stages involve some rather idiosyncratic procedures that are tied to the various ways different publishers code the information on the tapes, particularly since the tapes were not prepared with any consideration of later computational use. However, as will be apparent later, our second-stage normalization is a more general approach that we envision using with a variety of machine-readable dictionary resources. In what follows, we discuss the normalization of the Collins bilingual dictionaries.</Paragraph> <Paragraph position="6"> The first stage of normalization, picking out the headwords and the associated information, was mostly straightforward: the headwords are preceded by a unique font code. Exceptions to this straightforward process were driven by the principle that the end result of the process should be a direct access dictionary usable by both people and other systems. This involves definition of a file keyword which differs from that of the dictionary headword in two important ways: (1) each keyword must be unique, and (2) there should be a keyword for each spelling of a word, with the important exception of inflected forms predictable by morphological rules, as explained below. The first constraint assures that dictionary lookup for a word will get all of the information associated with that word. The second assures that a lookup operation will find a word where it first expects to find it, and not inside of some other entry.</Paragraph> <Paragraph position="7"> Three examples will suffice. In adherence to the first constraint, we combined the entries for the noun quack 1 and the verb quack 2 in the English-Italian Collins dictionary. In adherence to the second constraint, we created new entries for alternate spellings of words: the German-English headord Abtak(e)lung was made into two entries, Abtaklung and Abtakelung, the associated information was stored with just one of them, and cross-references to the other spelling were stored with both. Since the notation used for alternate spellings of this sort does not indicate which one is the more common usage, we stored the entry arbitrarily with the shorter form. A third example: where the German-English dictionary stores a number of compound words with the first element of the compound (e.g. Abend-: has -essen, -friede(n) -gesellschaft, etc.), we created full entries for all of them (e.g. Abendessen, Abendfriede, Abendgesellschaft, etc). In the process of reconstituting the compounds, however, we retained information on the location of the break to support possible later work on German noun compounds and to conform to another principle: &quot;do not throw away anything.&quot; An exception to the constraint that there should be separate entries for each spelling of a word, noted 220 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods before, is that we do not require creation of separate entries for those words which can be derived by morphological rules. For example, from German Lehrer(inJ), which is shorthand for Lehrer and Lehrerin j\[emale\], we created only an entry for Lehrer, leaving the (in 79 part for later segmentation. Identification of these exceptions is possible using heuristics and does not depend on having a system of morphological rules for the language of the dictionary. These heuristics are different for each dictionary.</Paragraph> <Paragraph position="8"> The program which accomplished stage one segmentation was written in a variant of PL/I using recursion to handle the interaction of morphological variants, alternate spellings, and reconstruction of compounds. The results have been used directly for several WordSmith applications.</Paragraph> <Paragraph position="9"> Once we had the dictionary entries segmented into DAM files, they were ready for the second stage of normalization: identifying and labelling the different types of data in the entries in a standard and consistent way. As we set out to create data bases from the first set of DAM files, it quickly became apparent that the usual data base architecture, characterized by a fixed number of fields for each entry and a fixed amount of space for each, would never work for dictionary entries, because there might be one value for some fields (like spelling), but multiple values for others (like part-of-speech, translation, etc.), or no value at all for still others (e.g. alternate spelling). Moreover, some information, like translation, is associated with values for other information, like part-of-speech. These facts lead to the observation that dictionary entries are best represented as shallow hierarchies of attribute-value pairs. We will return to this observation in the section on &quot;lexicai data bases&quot;.</Paragraph> <Paragraph position="10"> We have developed a parser, the goal of which is to convert each dictionary entry into a tree structure where at each node there is an attribute name and a value, which can be a simple value, a list of attribute-value pairs, or an entire subtree. Figure 1 contains the result of applying the parser to the entry for alert in the Collins English-Italian dictionary. The attributes in the figure should be familiar to users of bilingual dictionaries; they include hyphenation code (&quot;hyph'), undecoded pronunciations (&quot;pron&quot;), homograph numbers (&quot;hnum&quot;), part-of-speech (&quot;pos&quot;), translations (&quot;spel&quot; under &quot;xlat&quot;), collocations and examples in the source language (&quot;gloss&quot; under &quot;xmp&quot;), and translations of collocations and examples (&quot;tran&quot; under &quot;xmp&quot;). Figure 2 shows how the above-mentioned combination of the multiple entries for quack and other words brings about the creation of a &quot;superhom&quot; node. The superhom (&quot;superscript homograph&quot;) node serves to separate the multiple entries and preserve the original sense numbers. This is important because cross references in other entries may contain superhomograph numbers as well as sense numbers. Even in entries where there is only one sense, homograph, or superho-</Paragraph> <Paragraph position="12"> \[gloss=to ~ sb to the dangers of sth\] \[tran=mettere qn in guardia contro qc\]\]\]\]\] Figure 1. Normalized Collins English-Italian entry for alert. mograph, these attributes are retained to make access logic consistent with other entries.</Paragraph> <Paragraph position="13"> The design of the dictionary entry parser began with a familiar model for parsers of English sentences: a set of rules supported by a parsing engine. Two such models were available to us: a bottom-up, all-paths strategy offered by PLNLP (Langendoen and Barnett(to appear)), and a top-down depth-first approach offered by logic grammars written in Prolog (McCord(1987)).</Paragraph> <Paragraph position="14"> We have versions in PLNLP and VM/Prolog, both of which are basically adequate. The version which supported the work reported here is the Prolog version; however, we are concerned about future problems with the top-down approach when we have to process input that is corrupt. Usually for lexicographical processing, one should not have to worry about large amounts of ill-formed input, but the complex history of some of our tapes forces us to consider possible recovery strategies for text with missing or corrupt font-codes. At that point, it may be necessary to use the bottom-up approach and some recovery strategy analogous to parsefitting (cf. Jensen, et a1.(1983)).</Paragraph> <Paragraph position="15"> Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 221 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods</Paragraph> <Paragraph position="17"> Grammars for dictionary entries and grammars for natural language sentences differ in three important ways: (1) entries can be huge: some are longer than 5000 bytes; (2) there is a different definition of a token, and (3) a dictionary grammar can be to a large extent deterministic. A consequence of (1) is that it takes a large amount of storage to parse an entry. The Prolog version now runs on a 4-megabyte virtual machine under VM/CMS. The motivation for (2) comes from the fact that the entries consist of text interspersed with font-change codes, which are not required to be delimited with blanks. Whereas for ordinary language, the tokens are strings between blanks, for dictionary entries they are the strings of characters between a defined set of delimiters, usually font-change codes, but occasionally also semicolons, periods, and complex character strings. The delimiters are tokens as well. In view of (3), one might well ask why anyone would use a powerful tool like PLNLP or a logic grammar to accomplish something that does not need such powerful support.</Paragraph> <Paragraph position="18"> The answer is that we now have a dictionary parsing engine, a formalism for rules, and an easily modifiable set of rules, all achieved in a shorter development time than would have been accomplished with a conventi0nal programming language. Furthermore, stage two of the acquisition of new dictionaries now requires only writing, testing, and running a new set of grammar rules.</Paragraph> <Paragraph position="19"> The parsing engine and support code are written in VM/Pr01og. The rules, which may be different for each dictionary, are written using a formalism in the spirit of the &quot;definite clause grammars&quot; (DCG's) of Pereira and Warren (1980) and &quot;modular logic grammars&quot; (MLG's) of McCord (1986). A rule compiler translates the rules into Prolog code; this compiler owes much to the groundwork laid by McCord. Significant to McCord's work and necessary for ours here is the distinction between two different kinds of rules, one which produces a structural node (strong non-terminal), and one which produces an intermediate list of nodes (weak non-terminal). The dictionary entry rule compiler sports several new features required by dictionary entry parsing. These include an 'ignore' operator, which causes tokens in the input to be ignored by the automatic structure builder, an 'insert' operator which causes new elements to be inserted, and the efficient implementation of rules consisting entirely of optional elements. This generalized rule-driven parsing engine will be used to help develop entry grammars for all our dictionary resources; a small tool kit for this is already available, and other tools will be added as development progresses.</Paragraph> <Paragraph position="20"> We illustrate the rule formalism with the following two rules from the grammar for the Collins Italian-English dictionary:</Paragraph> <Paragraph position="22"> The first rule says that an unnumbered homograph consists of a number of optional elements, of which at least one must be present: pronunciation; choice of part-of-speech or verb-reflexive or none; morphological information; selection of perfect tense auxiliary; a certain kind of cross reference (marked with 'v'); a gloss; or a list of senses. The hom constituent is declared elsewhere as a strong nonterminal; a node is created for it, with daughter nodes as defined to the right of the arrow.</Paragraph> <Paragraph position="23"> Prolog backtracking conventions apply; the first successful parse of an entry is used, even if there might be more than one possibility. In the case of most optional constituents, the existence of the constituent is tried before its absence. In the case of the gloss in the first rule, however, the correct result is obtained if the absence of the gloss is tried first.</Paragraph> <Paragraph position="24"> 222 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods The second rule says that a list of numbered senses consists of a numbered sense and a numbered sense list (or nothing). The recursive property of this rule allows the assembling of any number of numbered senses. The senslist constituent is declared as a weak non-terminal, so that all the sub-constituents defined by the rule will appear as sister nodes at the same level, immediately dominated not by senslist, but by its parent, hom.</Paragraph> <Paragraph position="25"> The present version of the grammar has been run on parts of the Collins Italian-English and English-Italian dictionaries with a success rate of about 95%. Some of the entries were quite large. The high success rate is partly due to the consistency and discipline in the formatting of the Collins Italian dictionaries. The rules are still being improved. There are several reasons for the remaining 5% -- and, indeed, also for some of the ad hoc devices we have used to successfully parse some entries. Sometimes the primary clues to parsing, the font change codes, can be missing because two discrete data segments are contiguous and appear in the same font. Alternatively, the font changes can actually occur in the middle of a coherent data segment. For example, the translation for English queen bee is given as &quot;ape f regina inv&quot;. Here, &quot;f&quot; is the feminine gender marking for ape, and &quot;inv&quot; (i.e., &quot;invariable&quot;) indicates that the word regina remains invariable in the plural form.</Paragraph> <Paragraph position="26"> In this case, there is grammatical information about each of the pieces of the collocation. In other cases, the first piece of grammatical information attaches to the first word, and the second to the entire collocation. To take a second example, the locution &quot;x or y&quot; is often used for phrases in all languages. The problem is that both &quot;x&quot; and &quot;y&quot; can be gapped, or otherwise abbreviated, by deletion of words identical to those found on the other side of the &quot;or&quot;. An example is in the translation given for Italian si sta facendo buio: &quot;it is growing o getting dark&quot; (shorthand for &quot;it is growing dark o it is getting dark&quot;).</Paragraph> <Paragraph position="27"> For cases like these, we would ideally like to identify and extract all the information immediately. It would be nice to separate the grammatical information from ape regina, in the first case, and to reconstitute the translations it is growing dark and it is getting dark, in the second. However, we recognize that solving all such problems is a never-ending process and would indefinitely delay the time that any information is available for lexicological analysis. Consequently, we make a practical distinction between the acquisition and analysis phases of our work: information that is not easily decipherable using our entry parsing tools is left in its undeciphered form in the initial lexical data base that we create. Thus, for our examples, we should have the following attributes.</Paragraph> <Paragraph position="29"> Later work can be devoted to the detailed analysis of such complex attributes. The results of such analysis will be used to create more refined versions of the initial lexical data bases.</Paragraph> <Paragraph position="30"> 2.2. LEXICAL DATA BASES AND THE LEXICAL QUERY LANGUAGE.</Paragraph> <Paragraph position="31"> The lexical data base (LDB) facility is designed to store dictionary information and to provide fast, flexible, and convenient access to it. LDBs differ from DAM files in that the entries are structured into hierarchies of attribute-value pairs, such as those that result from the entry parsing described in the previous section. The lexical query language (LQL) allows users to query, modify, and format those entries. A prototype LDB/LQL system has been built as a testing ground for various concepts and algorithms and is currently in use, as described in sections 4 and 5. The final version of the facility is still under development.</Paragraph> <Paragraph position="32"> Dictionary entries are typically organized as shallow hierarchies of attribute-value pairs with a variable number of copies of certain nodes at each level. The hierarchical nature of entries stems from the fact that values can be entire subtrees. Thus, after processing by the dictionary entry parser described in the previous section, the entry for quack from the Collins English-Italian dictionary can be stored in an LDB as shown in Figure 3. In this hierarchy, the &quot;hdw&quot; terminal node contains the headword of the entry. Dictionary workers are familiar with the notion of hierarchically structured entries of this type.</Paragraph> <Paragraph position="33"> It is important to distinguish between the type of hierarchies found in dictionary entries and those, for example, that characterize the syntactic structure of English sentences. Both types of hierarchy are, in principle, of unbounded size. Dictionary entries achieve their unboundedness by iteration: a bilingual dictionary entry may have any number of homographs, which may, in turn, have an arbitrary number of senses, and so forth. Syntax trees achieve their unboundedness both by iteration (e.g., coordination) and by recursion (e.g., self-embedding). The observation that dictionary entry hierarchies may only iterate is useful for defining lexical data bases. All entries in an LDB can be characterized by a &quot;design&quot; which is stored once for the data base. The design is a single hierarchy which Serves as a &quot;grammar&quot; for any instance of an entry by naming the possible parent, siblings, and children of each node type that may occur. Actual LDBs are created by storing hierarchically formatted entries in a DAM file with the design stored as control information. Each entry is stored with its headword as key. Alternate access paths can be established by building indexes on other attributes. null Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 223 Roy J. Byrd, Nicoletta Caizolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods entry for quack.</Paragraph> <Paragraph position="34"> LQL allows the user to specify conditions on the attributes of LDB entries. Only those entries which satisfy all conditions become part of the query answer.</Paragraph> <Paragraph position="35"> Further, the user specifies which attributes of the successful entries remain in the answer and what their output format will be. The query is stated as entries in the nodes of a two-dimensional representation of an LDB's design. The query syntax is reminiscent of the syntax of the Query-by-Example (QBE) data base query language (Zloof(1974)). Example-elements, denoted by a leading underscore, are used to relate values of attributes in a query tree to conditions in a &quot;condition box&quot;, to display positions in an &quot;output box&quot;, and to attribute values in other dictionary entry trees.</Paragraph> <Paragraph position="36"> Part (a) of Figure 4 shows a query which will list all words which are both nouns and verbs in English together with their translations in the Collins English-Italian dictionary. The condition on the noun part-of-speech attribute is simple (it must be &quot;n&quot; for this data base) and is entered directly in the tree. The condition on the verb part of speech is more complex, and the example-element _vpos is used to relate those attribute values to the condition box, where they are tested for equality with either &quot;vi&quot; or &quot;vt&quot;. The example-elements _word, _ntran, and _vtran are used to map answers from the hierarchy into the output box where their eventual output format is schematically represented. Part (b) shows sample entries from the answer to this query when applied to the Collins English-Italian dictionary. Such a query might be useful, for example, in a study of the relation between homography in English and derivational morphology in Italian.</Paragraph> <Paragraph position="37"> LQL conditions specify tests to be performed on the values of attributes in the data tree to which nodes in the query tree may be mapped during query processing.</Paragraph> <Paragraph position="38"> 224 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods Terminal nodes may be tested using a variety of string and arithmetic operations. The current prototype implementation includes the built-in functions of the REXX programming language (IBM (1984)). Non-terminal nodes may only be tested for equality or inequality with other nodes of the same type. All nodes may have aggregate functions (e.g., count, maximum, minimum, etc.) applied to them and the results may either be tested in conditions or be output as part of the query answer. Nodes may also be tested for a null value (i.e., non-occurrence in a particular entry).</Paragraph> <Paragraph position="39"> As in the query tree itself, the output box may contain both constants and example-elements. The constants define boilerplate material used to label the variable instantiations of the example-elements, as in Figure 4. Such labelling is perhaps more useful when dictionary entries are presented one at a time, as in the WordSmith on-line dictionary system (Neff and Byrd(1987)).</Paragraph> <Paragraph position="40"> A realistic data base system must provide facilities for creating and maintaining data bases. LDB creation is usually a bulk operation and has been discussed in the previous section. LDB maintenance, on the other hand, can benefit from the flexibility provided by combining a powerful query processor with the capability to insert, delete, and update parts of the LDB entries. LQL offers this flexibility by providing the operators &quot;i.&quot; (for &quot;insert&quot;), &quot;d.&quot; (for &quot;delete), and &quot;u.&quot; (for &quot;update&quot;). These are also familiar Query-by-Example concepts and are described in a relational context in IBM(1978), and in the context of LDBs in Neff, et a1.(1988). LDB modification will not be discussed further in this paper.</Paragraph> <Paragraph position="41"> We envision that LQL will be available in multiple environments. Naturally as a stand-alone processor, it can be used to specify batch query and modification operations against entire lexical data bases. This processor and the flexibility with which it can produce results tailored to a specific problem make LQL a valuable tool for lexicological research. We also plan to use LQL as an entry formatting and filtering mechanism for the WordSmith on-line dictionary system.</Paragraph> <Paragraph position="42"> 3. TOOLS AND METHODOLOGIES.</Paragraph> <Paragraph position="43"> 3.1. TOOLS.</Paragraph> <Paragraph position="44"> In this subsection, we give an alphabetical listing of the major computational tools, other than the dictionary entry parser and LDB/LQL, at our disposal for doing lexicological analysis of our MRDs. The tools are described in more detail in the sections of the paper where their use in a certain application is presented.</Paragraph> <Paragraph position="45"> Further detail is available in the references. Minor tools used for individual applications are not shown here.</Paragraph> <Paragraph position="46"> The tools are in the form of a variety of programs written by different members of the Lexical Systems project. The programming languages used include IBM's EXEC2, REXX, PL/I, and VM/Prolog languages. What makes them coherent are the fact that we use the Dictionary Access Method and its utilities as a common interface to the data and that we share a common set of methods and goals when we apply them.</Paragraph> <Paragraph position="47"> DAM. The Dictionary Access Method (Byrd, et al.(1986b)) provides random and sequential access to dictionary entries associated with words which serve as search keys. The sequential access can be modified to observe &quot;correct&quot; alphabetical orders in any of several languages. An associated set of utilities makes the management of DAM files (including creation, updating, and compression) especially convenient.</Paragraph> <Paragraph position="48"> DICTUTIL. DICTUTIL is a menu-driven interface to the prototype implementation of the lexical data base system. It allows the analyst to create LDBs using output from the parsing of Collins bilingual dictionary entries and to execute simple queries against those LDBs. The output from a DICTUTIL query is stored as an index to the original LDB (i.e., as a set of valueheadword pairs).</Paragraph> <Paragraph position="49"> Filtering. Filtering, described in Chodorow, et a1.(1985), is a method for expanding a set of words each of which has a given property. It uses the hypernym relationship, produced by Head Finding, or the synonym relationship to find new words that are related only to words already in the set. Such new words are added to the set and the process repeats until convergence. See section 3.2, below.</Paragraph> <Paragraph position="50"> Head Finding. Head finding, also described in Chodorow, et a1.(1985) and in Calzolari(1984a), is a method for automatically discovering the hypernym (or genus term) for a word by analyzing its dictionary definition. It is based on the observation that, at least for nouns and verbs, the syntactic head(s) of a definition is(are) usually the hypernym(s) of the word being defined. See section 3.2, below.</Paragraph> <Paragraph position="51"> Matrix Building. The Matrix Building program takes ordered pairs of words that bear a given relationship to one another (e.g., X is a synonym of Y) and constructs a matrix in which each X word is represented by a row, each Y word is represented by a column, and the relationship between them is indicated in the cell that is formed by the XY intersection. Synonym matrices are used for clustering synonyms into senses. Translation matrices (where X is a translation of Y) are used for transferring and analyzing sense information using bi-lingual dictionaries. See sections 5.1 and 5.2, below. Sprouting. Sprouting, described in Chodorow, et a1.(1985), is a method for exploring the hypernym, hyponym, and synonym relations. It finds words related to some initial seed by computing the transitive closure over the relation (i.e., &quot;sprouting a tree&quot;) beginning from the seed. See 4.1, below.</Paragraph> <Paragraph position="52"> TUPLES. TUPLES (Byrd(1986b)) is a text analysis system for finding frequent words and phrases in ordinary text. It uses UDICT to provide lemma normalization and base normalization as it compares strings from Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 225 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods the text. TUPLES is used in the analysis of dictionary definitions for locating and counting defining formulae and presenting them in a key-word-in-context format.</Paragraph> <Paragraph position="53"> UDICT. UDICT (Byrd(1986a)) is a computerized lexicon system. It supplies a rich variety of lexical information to natural language processing applications. Of particular interest for lexicography is the fact that UDICT performs morphological analysis of words having inflectional and derivational affixes as well as for compound words and multi-word entries. It also has a facility for producing correctly inflected form of English words. UDICT is a component of TUPLES and WordSmith. null WordSmith. WordSmith is an on-line dictionary system, described in Neff and Byrd(1987), which allows flexible access to dictionaries stored as DAM files and lexical data bases. Although not explicitly mentioned in the remainder of the paper, it is an important reference tool for all of our research. Among its capabilities, WordSmith provides access to: * definitions, synonyms, and etymologies from Web- null ster's Seventh (Merriam(1963)), * pronunciations from Webster's Seventh and rhymes based on them, * definitions and grammatical information from LDOCE (Longman(1978)), * synonyms from the Collins Thesaurus (Collins(1984)), * entries from the Collins bilingual dictionaries for English/Italian, English/French, English/Spanish, and English/German (Collins(1971, 1978, 1980, 1981)).</Paragraph> <Paragraph position="54"> 3.2. METHODOLOGIES.</Paragraph> <Paragraph position="55"> At the beginning of our work, we dealt with words as undifferentiated wholes. That is, we only stored lexical information for a word when it applied to all of the word's senses for a given part of speech. Even though this practice made us somewhat uncomfortable, it allowed us to make significant progress in dictionary analysis. The primary reason, as shown in Figure 5, is that most words in the dictionaries we analyze are monosemous; that is, they have just one sense. Of the approximately 60,000 headwords in Webster's Seventh having fewer than ten senses, almost two-thirds of them (over 38,000) are monosemous. Further analysis, not given here, shows that, even among words that have multiple parts-of-speech, each part-of-speech is also typically monosemous. This further increases the proportion of the dictionary entries that we can handle with our earliest techniques. Analysis of Zingarelli(1971) reveals that the facts for Italian are analogous to those for English: over all parts-of-speech, 61.7 percent of the Italian entries were monosemous; 38.3 percent were polysemous. Further details of the Italian study are presented in Calzolari(1980).</Paragraph> <Paragraph position="56"> These statistics explain why MRD-based methodologies such as head-finding, sprouting, and filtering are as productive as they are, even though they don't handle polysemy correctly. They are described in section 3.2.1. Despite these encouraging statistics, however, it is ultimately necessary to be able to deal correctly with the multiple senses of polysemous words.</Paragraph> <Paragraph position="57"> The need is especially acute because the most frequently used words tend to also be the most polysemous. Two specific problems associated with polysemy in MRDs are the mapping problem and the addenda problem. Section 3.3 describes these problems and offers solutions for them.</Paragraph> <Paragraph position="58"> 3.2.1. HEAD FINDING, SPROUTING, AND FILTERING.</Paragraph> <Paragraph position="59"> One goal of our research is to extract semantic and syntactic information from standard dictionary definitions for use in constructing lexicons for natural language processing systems. Dictionaries are rich sources of detailed information, but in order to use the information for natural language processing, it must be organized systematically.</Paragraph> <Paragraph position="60"> Amsler(1980) demonstrates that additional structure can be imposed upon a dictionary by making certain assumptions about the ways in which definitions are constructed. Foremost among these assumptions is that definitions consist of one or more &quot;genus&quot; terms, which identify superordinate categories of the defined word, and &quot;differentia&quot; which distinguish this instance of the superordinate categories from other instances. By manually extracting and disambiguating genus terms for a pocket dictionary, Amsler demonstrated the feasibility of generating semantic hierarchies.</Paragraph> <Paragraph position="61"> It has been our goal to automate the genus extraction and disambiguation processes so that hierarchies can be generated from full-sized dictionaries, such as Webster's Seventh New Collegiate Dictionary. These efforts are described in detail in Chodorow, et al. (1985). They begin with Head Finding.</Paragraph> <Paragraph position="62"> In the definition of car as &quot;a vehicle moving on wheels&quot;, the word vehicle serves as the genus term, while &quot;moving on wheels&quot; differentiates cars from some other types of vehicles. Taken as a group, all of the word/genus pairs contained in a normal dictionary for words of a given part-of-speech form what Amsler(1980) calls a &quot;tangled hierarchy&quot;. In this hierarchy, each word constitutes a node whose subordinate 226 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods nodes are words for which it serves as a genus term.</Paragraph> <Paragraph position="63"> The words at those subordinate nodes are called the word's &quot;hyponyms&quot;. Similarly, the words at the superordinate nodes for a given word are the genus terms for the various sense definitions of that word. These are called the given word's &quot;hypernyms&quot;. Because words are polysemous any word may have multiple hypernyms; hence the hierarchy is &quot;tangled&quot;.</Paragraph> <Paragraph position="64"> Our automated approach to finding the genus terms in definitions is based on the observation that the genus term for verb and noun definitions is typically, though not always, the syntactic head of the defining phrase.</Paragraph> <Paragraph position="65"> This reduces the task to that of finding the heads of verb phrases and noun phrases. The syntax of the verb phrase used in verb definitions makes it possible to always locate its head with a simple heuristic: the head is the single verb following the word to, or if there is a conjunction of verbs following to, then they are all heads.</Paragraph> <Paragraph position="66"> Head finding is much more complex for noun definitions because of their greater variety, but we have taken advantage of the fact that dictionary definitions are written in a special and predictable style to develop a heuristic procedure for finding the head. The procedure can be described briefly as follows. First the substring which must contain the head is found. This substring is bounded on the left by a word which obligatorily appears in prenominal position: a, an, the, its, two, three ..... twelve, first, second, and so forth. It is bounded on the right by a word or sequence that can only appear in postnominal position, such as a relative pronoun, a preposition, or a present participle following a noun.</Paragraph> <Paragraph position="67"> Once the substring is isolated, the search for the head begins. Typically, but not always, it is the rightmost noun in the substring. If however, the substring contains a conjunction, each conjunct is processed separately, and multiple heads may result. If the word found belongs to a small class of &quot;empty heads&quot; (words like one, any, kind, class, manner, family, race, group, complex, etc.) and is followed by of, then the string following of is reprocessed in an effort to locate additional heads. The accuracy of head-finding for nouns was approximately 98 percent, based on a random sample of the output.</Paragraph> <Paragraph position="68"> The word defined and the head of its definition are the raw materials for two semi-automatic processes which make explicit an important part of the lexical organization and featural information that are held implicit in dictionaries. The first of these is sprouting.' Sprouting, which derives its name from the action of growing a semantic tree from a specified root, uses the results of head-finding organized into a &quot;hyponym index&quot;, in which each word that was used as a genus term is associated with all of its hyponyms. Thus, vehicle would have an entry in the index which reads (in part): vehicle: ambulance . . . bicycle . . . car...</Paragraph> <Paragraph position="69"> tanker . . .</Paragraph> <Paragraph position="70"> For a given part-of-speech, the hyponym index needs to be built only once.</Paragraph> <Paragraph position="71"> When invoking the SPROUT program, the user selects a root from which a semantic tree is to be grown. The system then computes the transitive closure over the hyponym index, beginning at the chosen root. In effect, for each new word (including the root), all of its hyponyms are added to the tree. This operation is applied recursively, until no further new words are found.</Paragraph> <Paragraph position="72"> The interactiveness of the sprouting process results from the fact that the user is consulted for each new word. If the user decides that the word does not belong to the tree being grown, it can be pruned. These pruning decisions result in the disambiguation of the tree, though clearly not by means of an automatic process.</Paragraph> <Paragraph position="73"> The output of a sprouting session, then, is a disambiguated tree extracted from the tangled hierarchy represented by the hyponym index. The words it contains all have at least one sense which bears the property for which the root was originally selected. It is important to note that any serious use of sprouting to locate all words bearing a particular semantic feature must involve the careful selection and use of several roots, because of the variety of genus terms employed by the Webster's lexicographers. This will be quite obvious in the case studies cited below.</Paragraph> <Paragraph position="74"> Filtering, like sprouting, results in lists of words bearing a certain property (e.g., \[+human\]). Unlike sprouting, however, filtering only picks up words all of whose senses have the property. It is based on a hypernym index (the inversion of the hyponym index), in which each word is listed with its hypernyms, as in the example given here: vehicle: agent equipment means medium The filtering process begins with a &quot;seed filter&quot; consisting of an initial set of words all of whose senses bear some required property. The seed filter may be obtained in any manner that is convenient. Several approaches to setting up a seed filter are discussed in the case studies that follow. Given the filter, the system uses it to evaluate all of the words in the hypernym index. Any words, all of whose hypernyms are already in the filter, become candidates for inclusion in the filter during the next pass. The user is consulted for each candidate, and may accept or reject it. Finally, all accepted words are added to the filter, and the process is repeated until it converges.</Paragraph> <Paragraph position="75"> Like sprouting, filtering produces a list of words having some desired property. In the case of filtering, however, the resulting words have the property in all of their senses.</Paragraph> <Paragraph position="76"> 3.3. MULTI-DICTIONARY LEXICOLOGY.</Paragraph> <Paragraph position="77"> When lexicographers create dictionaries, they identify varying numbers of senses for words based on such Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 227 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods things as dictionary size, editorial policy, available citations, and intended audience. Computer scientists who create or add to a dictionary for supporting natural language processing applications have other criteria which guide the sense distinctions to be made. It is not to be expected that different MRDs will agree on the numbers or orders of senses for the words they contain. The disagreement, however, raises problems for computational lexicology, two of which are discussed in this section.</Paragraph> <Paragraph position="78"> 3.3.1. THE MAPPING PROBLEM.</Paragraph> <Paragraph position="79"> When processing multiple MRDs, one encounters what we have termed the &quot;mapping problem&quot;. The problem is to map the senses given for a word in one dictionary onto the senses given for the same word in another dictionary. We see a mapping as a symmetric binary relation between (word senses in) two dictionaries for the same language.</Paragraph> <Paragraph position="80"> A solution to the mapping problem is necessary in order for us to combine information in different published dictionaries. For example we might want to relate grammatical information from the Longman Dictionary of Contemporary English with definitions from Webster's Seventh. Having correct mappings will also allow us to transfer information from one dictionary to another when we are building new computerized dictionaries such as UDICT. We note that some computerized dictionaries will have artificially created sets of senses -- possibly motivated by discriminations required by the application being served. Our mapping strategy must be capable of handling these also.</Paragraph> <Paragraph position="81"> Figure 6 contains a general picture of the mapping relation among 3 different dictionaries (Webster's Seventh, Collins English-Italian, and Italian monolingual) together with evidence for establishing a specific mapping. In the figure, the English word E 1 has three senses in Webster's Seventh, two known to refer to humans and</Paragraph> <Paragraph position="83"> one to an instrument. The Collins English-Italian dictionary has two senses for El; with two Italian words I1 and 12 given as translations of the respective senses. Finally a monolingual Italian dictionary contains the information that I1 is human and that 12 is an instrument. Using techniques and representations to be described below, we can establish the mapping shown between Webster's Seventh and Collins English-Italian; the first two senses in Webster's map to the first sense in Collins and the third in Webster's maps to the second in Collins.</Paragraph> <Paragraph position="84"> Solving the mapping problem involves two kinds of activities. First, a representation must be chosen for storing the mappings as they emerge from various procedures intended to find them. This decision is merely one of choosing permanent file structure for the binary relations. (Note that we reject the alternative of discovering mappings dynamically when they are needed because (1) no single procedure can be devised to find them all and (2) most procedures for finding any mappings would be too expensive for dynamic application.) In the second activity, we must devise and run procedures for obtaining the detailed mappings. In general, multiple procedures will be required in order to fully cover the words in the two dictionaries to be mapped.</Paragraph> <Paragraph position="85"> Any file system that supports random access of information keyed by words will be suitable for storing the mappings. DAM will be used for the mapping relations built at IBM Research. In the DAM files, headwords from the two dictionaries being mapped will serve as keys.</Paragraph> <Paragraph position="86"> The information stored for each headword can be a set of binary mappings. For example, we give here the mapping for the word E1 in Webster's Seventh and Collins English-Italian shown in Figure 6: Collins's sense 2. Here the binary mappings are separated by semicolons. Each mapping consists of two lists of senses separated by colons. The leftmost list is for Webster's Seventh, the rightmost for Collins English-Italian. Since the mappings are symmetric, the assignment of left or right to a particular dictionary is arbitrary. Each list, in turn, is a comma-delimited set of sense numbers. A list may be null if one dictionary excludes a sense contained in the other. (It is important to note, however, that the success of the mapping strategy depends on the MRDs involved having fairly complete coverage of the language. Arbitrary omission of word senses will, in general, lead to incorrect results.) With respect to the number of non-null word senses in the two entries for a given word, there are four possible types of mappings, as shown in the table in Figure 7.</Paragraph> <Paragraph position="87"> good as having a single translation. Likewise, if a single translation is polysemous in Italian but all senses bear the relevant feature, then the observations still hold. Further generalizations should yield a useful procedure for instantiating (part of) the desired mapping.</Paragraph> <Paragraph position="88"> Many such procedures, possibly augmented by some manual labor, will be required to build an entire mapping. In particular, we expect that for highly polysemous and common words (such as run, go, turn, thing, make, be, etc., in English) the mappings will have to be done by hand. Once obtained, however, the mapping will be permanent and can be used for a variety of purposes, some of which will be discussed in sections 4 and 5 of this paper. 3.3.2. THE ADDENDA PROBLEM.</Paragraph> <Paragraph position="89"> Sense mapping is not the only aspect of work with computerized dictionaries where the ability to identify Type # senses # senses on left on right entry in mapping file</Paragraph> <Paragraph position="91"> We can imagine very simple strategies for populating the mapping file for mappings of types 1,2 and 3. We simply build the mapping entry shown in the fourth column of the figure. Given our previous observation that most entries in a typical dictionary are monosemous, these strategies will account for a significant portion of the mapping file entries.</Paragraph> <Paragraph position="92"> Mappings of type 4 will require more involved strategies. As an example of one such strategy, consider the example in Figure 6. If I1 is \[+human\] and is the only translation given for the first sense of E1 in Collins English-Italian, then the first two senses of E1 in Webster's may be mapped to the first sense of E1 in Collins. Likewise, if I2 is \[+instr\] and is the only translation given for the second sense of E1 in Collins, then we may assume that the third sense of E1 in Webster's (which is \[+instr\]) can be mapped to the second sense of E1 in Collins. A procedure based on these observations will yield the mapping shown before, namely: Note that the procedure will include generalizations of some aspects of the observations made above. For example, if Collins gives multiple translations for a single sense of a word, and if all are \[+human\] that is as word senses is critical. Activities which develop dictionary information--either by dictionary and text analysis or by transferring the information from other dictionaries-must ultimately have word senses as targets for that information. The information to be associated with the target dictionary is generally represented either as features, or as relations (i.e., attribute-value pairs).</Paragraph> <Paragraph position="93"> The dictionary to which the information is to be attached may either be a published dictionary, like Webster's Seventh, or a dictionary intended for computer use, such as UDICT. In the former case, the senses are those established by the lexicographers. In the latter case, the senses are those which meet the needs of the application(s) which the dictionary serves.</Paragraph> <Paragraph position="94"> We will also need to decide whether the new information should actually be added to the existing file for the target dictionary or, rather, whether it should be kept in a separate &quot;addendum&quot; to the base file, leaving the base file unchanged. Ignoring, for the moment, the first possibility, we propose the following strategy for adding new information as an addendum to an existing base dictionary.</Paragraph> <Paragraph position="95"> The information must be addressable by word and sense number, where the sense numbers are those given in the base dictionary. However, the need to access Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 229 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods information by sense number, independently of the word, will never arise. Consequently, a DAM file in which headwords from the base file are keys, and in which sense numbers are used to organize the data records, is a suitable implementation. In particular, the file can be organized as an LDB whose entries have the following structure: To take a concrete example, suppose that analysis techniques described in this paper have been used to determine that the first sense of the noun mother in Webster's Seventh is \[+animate\] and takes \[+animate\] nouns as typical objects (the mother of my puppies is grammatical but *the mother of my buggy is not).</Paragraph> <Paragraph position="96"> Further analysis will show that the third sense of mother (&quot;maternal tenderness or affection&quot;) is \[+abstract\] and \[+mass\]. The LDB entry for mother in the addenda file for Webster's Seventh will have the following structure: With join operations in LQL, we will have a flexible means of querying, formatting, and otherwise combining the new information with that from the base dictionary. null</Paragraph> </Section> <Section position="4" start_page="0" end_page="0" type="metho"> <SectionTitle> 4. APPLIED COMPUTATIONAL LEXICOLOGY: MONOLINGUAL DICTIONARIES. </SectionTitle> <Paragraph position="0"> As we pursue our goal of building computerized dictionaries containing extensive lexical information, we use our data bases, tools, and methods to perform a variety of analyses. This section and section 5 present samples of those analyses. The variety will be evident in the different types of results sought and methods used. It is also reflected in the fact that some of these analyses have been completed, whereas others are still in progress. null</Paragraph> </Section> <Section position="5" start_page="0" end_page="0" type="metho"> <SectionTitle> 4.1. EXTRACTING SYNTACTIC AND SEMANTIC </SectionTitle> <Paragraph position="0"> INFORMATION.</Paragraph> <Paragraph position="1"> We have established a list of semantic and syntactic information which we intend to obtain for the purpose of building dictionaries suitable for natural language processing. For nouns, we have chosen information which is primarily motivated by selectional restrictions, such as \[+human\]/\[+nonhuman\], \[+male\]/\[+female\], \[+abstract\]/\[+concrete\]. Also for nouns, we are identifying attributes such as &quot;made_of&quot; and &quot;set_of&quot;, and features such as \[+unit\] (explained below)* For verbs, we are currently extracting information such as \[+thatcomp\] (takes a that complementizer), \[+active\]/ \[+stative\], \[+reflexive\], \[+reciprocal\] and \[selects for a subject of X type\]. Note that some of this information can be represented in terms of binary features, whereas other information should be represented as attribute-value pairs.</Paragraph> <Paragraph position="2"> There appears to be independent consensus in the computational literature for choosing information such as this. For example, Ingria (1986) refers to a set of lexical features necessary for the construction of large lexicons both for text analysis and generation systems.</Paragraph> <Paragraph position="3"> McCord(1987) refers to the features \[+human\]/ \[+nonhuman\], \[+male\]/\[+female\], and \[+animal\]/ \[+nonanimal\]. For McCord, semantic constraints can be exercised by operations on hierarchies of features with implicational relations. Dahlgren and McDowell(1986) present a system for reasoning about natural language texts. Among the primitive concepts in their system are ones we too have identified, such as PROCESS, ACTIVITY, MOTION, ANIMATE, IN-ANIMATE, INDIVIDUAL, COLLECTIVE, and so on.</Paragraph> <Paragraph position="4"> We believe that some of the same lexical information may be required for different languages. So far we have preliminarily examined Italian and English; the information we have identified for each language independently turns out to be of mutual interest and utility. This suggests to us that as we examine other languages, we might find a core set of syntactic and semantic attributes of universal linguistic interest which should be represented in the computational lexicon across languages. 4.1.1. \[+MALE\] AND \[+FEMALE\] NOUNS.</Paragraph> <Paragraph position="5"> Marking nouns with a feature reflecting semantic gender in English is motivated by both syntactic and semantic facts. Syntactically, processes like pronominalization are dependent on gender in English. Semantically, certain verbs, such as impregnate, select for \[+male\] or \[+female\] arguments. The criteria for defining \[+male\] and \[+female\] are relatively straightforward. We chose the option of two features because this is not a binary 230 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods distinction, e.g. happiness is both \[-male\] and \[-female\], whereas author could be either \[+male\] or \[+female\].</Paragraph> <Paragraph position="6"> We used three methodologies to extract nouns that are candidates for marking with the gender features.</Paragraph> <Paragraph position="7"> The initial list was extracted from Longman using the semantic codes which indicate &quot;male human&quot;, &quot;female human&quot;, &quot;male animal&quot;, and &quot;female animal&quot;. We then ran sprouting and filtering to enlarge the lists. The preliminary lists for the \[+male\] and \[+female\] features using sprouting and filtering were processed by Barbara Ann Kipfer (see Kipfer(1985)). The lists so obtained consisted of 732 entries for the \[+male\] feature and 519 for the \[+female\] feature.</Paragraph> <Paragraph position="8"> In addition to filtering, we searched entries containing the key words man, boy, husband, son, and male for the candidate \[+male\] words, and woman, girl, wife, daughter, and female for potential \[+female\] words.</Paragraph> <Paragraph position="9"> The simple key-word search technique turned up many items, particularly from the domain of botany, that were not necessarily \[+male\] or \[+female\]. This is not unexpected, and indeed confirms the motivation for developing an intelligent system such as filtering, rather than a less clever system such as the simple key-word search. However, the key-word search provided us with items that filtering did not catch. In particular, multi-word entries, such as man Friday or office girl were not picked up by filtering, but were picked up by the key-word search. Also problematic were noun compounds within a definition, such as: The heads of these definitions are aviator or adherent, rather than woman. Yet the words being defined belong to taxonomies other than the ones given by the genus term of their definitions. We need to explore other syntactic cues (such as noun compounding) to membership in multiple taxonomies.</Paragraph> <Paragraph position="10"> Finally, we extracted a list of nouns ending in the gender-marked suffixes in English, such as -ess, -ette, or -ix, for \[+female\] and -man for \[+male\]. The number of endings for \[+female\] is greater since this is the marked case in English, as seen in cases like president/woman president/ *man president or lawyer/ woman lawyer/ *man lawyer. This method accounts for about 15% of the words on the final list.</Paragraph> <Paragraph position="11"> Results. The lists we have for the features \[+male\] and \[+female\] consist of nouns which have the feature in all of its senses. In the original unprocessed lists from sprouting and filtering, there were 519 \[+female\] nouns, and 732 \[+male\]. For \[+female\], there were 230 entries eliminated from the filtered list, 85 added from the key word search, and 43 added from the backwards search for morphologically feminine nouns, giving a final total of 417 nouns marked \[+female\]. For \[+male\], there were 540 entries eliminated from the filtered list, and 36 added from the key word search, giving a final total of 228 nouns marked \[+male\]. Notice that many more male nouns had to be eliminated. The reason for this is that the male case is unmarked in English, and often serves as the generic term referring to both the female and male version of a type. For example, author can be used for men or women, whereas authoress always refers to women.</Paragraph> <Paragraph position="12"> The issue of sense discrimination plays a important role in marking words in the lexicon. Many nouns on our lists were likely to be either inherently \[+male\] or inherently \[+female\]. These words are of two types.</Paragraph> <Paragraph position="13"> The first includes words which have several senses not all of which possess the required feature. For example, among the definitions for king in Webster's Seventh are: The last three definitions are clearly inanimate, so marking a chess piece, card, or checker as \[+male\] is incorrect. However, in a certain sense, the meaning of king is intuitively male. Since we have at the present time no way to distinguish senses, we cannot mark certain of the senses \[+male\] while leaving the other senses neutral. A particular application, however, might have criteria for ignoring certain word senses. For example, in a context dealing with chess, king might always bear the features \[-human\] and \[-male\]. Most other applications might ignore the \[-human\] sense of king.</Paragraph> <Paragraph position="14"> The other words which are likely to bear a particular feature are those which are culturally or traditionally associated with a particular attribute. Such words are prostitute (a male prostitute is usually specified as such), major-general, Pope and beautician. This information is usually not explicitly stated in a definition. At the moment, we have no mechanism to assign a feature &quot;likely to be X&quot;, but this is in our future plans. Many of the words that appeared on our initial lists were adjectives that are used to refer to nouns which are likely to be either \[+male\] or \[+female\], such as petite, husky, and platinum-blonde. If these adjectives were marked as likely to select for nouns of a particular type, then this marking could in turn be used to identify nouns of that type. For example, if an adjective like petite were marked with a feature \[likely to select for a female noun\], then texts could be searched to pick out candidate \[+female\] nouns based on the presence of \[+female\] adjectives. This could be the basis for another methodology for identifying nouns possessing a certain semantic or syntactic feature. So far our lexical Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 231 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods information is confined to inherent features for nouns but a logical next step is to also mark selectional features for adjectives.</Paragraph> <Paragraph position="15"> 4.1.2. \[+UNIT\] NOUNS.</Paragraph> <Paragraph position="16"> We have marked nouns such as dollar, peso, mile, and acre with the feature \[+unit\]. The motivation for this arose from the following syntactic facts: Verb Agreement. It is necessary to &quot;break the rules&quot; of verb agreement when using unit nouns. For example, (1) Three dollars is too much to pay.</Paragraph> <Paragraph position="17"> (2)*Three dollars are too much to pay.</Paragraph> <Paragraph position="18"> (3) Four acres is too large for us.</Paragraph> <Paragraph position="19"> (4) *Four acres are too large for us.</Paragraph> <Paragraph position="20"> Agreement within a Noun Phrase. Usually a plural number marker requires a plural noun within the same noun phrase, such as &quot;three dollars&quot; or &quot;four acres.&quot; However, within a noun compound, this rule does not apply, even when the noun is clearly plural: There appear to be some differences between unit nouns and other count nouns. It seems to be that unit nouns obligatorily require a verb inflected for single number, whereas other count nouns optionally permit a singular or plural verb as shown in (9)-(10).</Paragraph> <Paragraph position="21"> (9) Four books is too much to read in one night.</Paragraph> <Paragraph position="22"> (10) Four books are too much to read in one night.</Paragraph> <Paragraph position="23"> The methods we used to identify dictionary entries which should be marked with the \[+unit\] feature are similar to those used for \[+male\] and \[+female\]. We obtained two lists, one from sprouting and filtering, and another from the key-word search program.</Paragraph> <Paragraph position="24"> Results. The number of unit nouns on the original filtered Longman list was 470. The key-word search technique turned up 527 more candidate words. Subtracting the 153 words common to both lists, there remained 844 words to be checked. Half of these passed the criteria in all senses, so a total of 422 words are marked with the \[+unit\] feature. In addition, a subset of unit nouns (n=128) were marked with a feature \[+currency\], of which 26 have irregular plural forms, yielding a total of 154 tokens marked. This was an unexpected but useful result particularly for judging the grammaticality of constructions like: Rice is four dollars a pound.</Paragraph> <Paragraph position="25"> 4.1.3. ACTIVE AND STATIVE VERBS.</Paragraph> <Paragraph position="26"> Markowitz et al. (1986) present linguistic motivations for extracting an &quot;active&quot; and &quot;stative&quot; feature for verbs, and suggest methods for using MRD's as sources for finding active/stative information for verbs. We decided to try to carry out their suggestions using our analysis tools and our morphological analyzer on our dictionary resources. Our attempt was both a success and a failure. The success arises from the design and use of our tools. We achieved clear results, although the results were not what we had expected. The disappointment arises from the fact that the active/stative distinction, it turns out, is not a lexical property at all, but rather a property of sentences. Our tools are geared towards lexical properties and are not designed to extract sentence or discourse properties. It is unclear how or whether to represent such attributes as lexical features.</Paragraph> <Paragraph position="27"> Linguistic Motivations. The standard division between Stative (or Status) verbs and Process (or active or dynamic) verbs revolves around aspectual constraints.</Paragraph> <Paragraph position="28"> The classic test is the ability to take progressive and imperative aspect as shown in the following examples: (I 1) The boy classifies the butterflies.</Paragraph> <Paragraph position="29"> (12) The boy is classifying the butterflies.</Paragraph> <Paragraph position="30"> (13) Classify the butterflies! (14) The boy resembles his father.</Paragraph> <Paragraph position="31"> (15) *The boy is resembling his father.</Paragraph> <Paragraph position="32"> (16) *Resemble your father! The verb classify is a process or dynamic verb. The progressive aspect is said to be possible with these verbs because they indicate a condition which can change over time. Similarly, the imperative is possible because one can order someone to do something in order to change the situation or condition. Other examples of dynamic verbs are ask, grow, hurt, arrive, and jump. Stative verbs, on the other hand, do not indicate a condition which can be changed by an action. As illustrated in (14)-(16), someone either resembles someone else, or he doesn't. Also, it is meaningless to order someone to resemble someone else. &quot;Resemblance&quot; is a condition, and not a changeable state. Process verbs are said to outnumber stative verbs, not only by type but also by token. Quirk and Greenbaum(1972) identify five types of process verbs, but only two types of stative verbs.</Paragraph> <Paragraph position="33"> Many stative verbs can be used as process verbs with certain changes in meaning. These meaning shifts might be reflected in a dictionary by different sense numbers since the fundamental reason for sense division is semantic. Consider the stative use of the perception verb hear: (18) I hear the music on the radio.</Paragraph> <Paragraph position="34"> (19) * I am hearing the music on the radio.</Paragraph> <Paragraph position="35"> (20) The judge is hearing your case first.</Paragraph> <Paragraph position="36"> As shown in (19) the progressive is ungrammatical when the verb hear is used with a stative meaning, whereas (20) is fully grammatical since the interpretation of hear is as a process. The problem of identifying both an active and a stative sense of a verb is related to the sense discrimination problem since a shift from active to stative sense has syntactic implications.</Paragraph> <Paragraph position="37"> Methodology and Tools. First, we obtained the list of noun hyponyms for &quot;act&quot; and &quot;state&quot; from our hyponym file. We then analyzed these nouns using a modified version of the morphological analyzer con232 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods strained to give the outermost verb. The expected result was that the outermost verb bases of these nouns would exhibit the active and stative properties depending on the source hypernym. We give some examples of definitions which have the genus act or state. Observe that resemblance which is derived from the stative verb resemble has state as one of as its genus term; abbreviation has act; and abatement has both act and state RESEMBLANCE (n) the quality or state of resembling : . . .</Paragraph> <Paragraph position="38"> ABBREVIATION (n) the act or result of abbreviating : . . .</Paragraph> <Paragraph position="39"> ABATEMENT (n) the act or process of abating : the state of being abated . . .</Paragraph> <Paragraph position="40"> The results of morphological analysis on these sample nouns yield resemble, abbreviate, and abate as their verb bases.</Paragraph> <Paragraph position="41"> Our processing of Webster's Seventh showed that there were 2185 words that had the noun state as a hypernym, for example, abatement, abidance, ability, and abnormality. The number of nouns having act for the hypernym was just slightly fewer at 2115, for example, abatement, abbreviation, aberration, and abidance. Not all have verbs bases, e.g. aberration, and abnormality. Observe that since many nouns have both hypernyms, the corresponding verbs end up on both lists, e.g. for abatement. The results are shown in as input each verb to be tested. Test sentences are constructed and the user is asked to respond whether a sentence is acceptable, unacceptable, or questionable.</Paragraph> <Paragraph position="42"> There is also opportunity for comments. The results are placed in a matrix containing the collected judgements for a list of verbs.</Paragraph> <Paragraph position="43"> Before running our lists of derived verbs through the sentence frame tool, we decided to test it with a verbs which have been said to be undeniably stative and active. We took seventy stative verbs and forty action verbs from Joos (1968) and Quirk and Greenbaum (1972).</Paragraph> <Paragraph position="44"> To illustrate, given the sample list of three verbs in (24), frame tests were automatically generated for each verb in turn, as shown in (25)-(27). The list in (24) has three verbs: one said to be stative (resemble), one active (throw), and one process (feel). Results are tabulated for each response, and given in (28).</Paragraph> <Paragraph position="45"> The table given in (28) shows the results of applying the frame tests to the verbs given in (24): The expected result if the distinction were as it is said to be is given in (29).</Paragraph> <Paragraph position="46"> Once we derived the verbs, we were then ready to test the original hypothesis. If the hypothesis is correct, verbs underlying nouns which have state as a hypernym are likely to be stative. Similarly, verbs underlying nouns which have act as a hypernym are likely to be active.</Paragraph> <Paragraph position="47"> We constructed the following syntactic tests from Joos (1968) and Quirk and Greenbaum (1972) which are said to characterize stativity: (1) the simple present progressive, (2) the present progressive with the adverbial modifier &quot;more and more each day&quot; in order to test a continual or process reading, and (3) the imperative. We built a sentence frame generating tool which takes (28) Results of Testing Verbs for Stativity: TEST 1 2 3 resemble .......... - + throw .............. + - + feel .................. + + + (29) Ideal Pattern TEST 1 2 3 active verb ....... + + + stative verb .......</Paragraph> <Paragraph position="48"> Notice that the patterns in (28) do not conform to the expected patterns in (29).</Paragraph> <Paragraph position="49"> A more complete set of responses using a random Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 233 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods subset of the verbs taken from Joos (1968) and Quirk and Greenbaum (1972), again failed to show any clear patterns or any clear distinctions among sets of verbs which are supposedly active and stative.</Paragraph> <Paragraph position="50"> Despite the fact that textbooks set out explicit distinctions for verb types, there appear to be no significant clear-cut core set of verbs in each category. Rather, there appear to be situations in which a given verb has a stative reading and other situations in which the same verb has an active reading. What is worse, these &quot;situations&quot; can be either semantically or syntactically conditioned.</Paragraph> <Paragraph position="51"> Choice of Auxiliary Verb in Italian. We next explored the possibility of a correspondence between verbs which tend to be active in English and those which select for the auxiliary avere in Italian, and similarly for the stative verbs in English and those requiring essere. Grammar books often state that in Romance languages, the choice of auxiliary verb depends on whether the verb is stative or active. We performed two tests of this hypothesis. The first was to extract some verbs from the Collins English-Italian dictionary which were labelled as taking essere &quot;to be&quot;. Verbs are not marked for avere &quot;to have&quot; since it is the unmarked case. We then passed these verbs through a frame test, identical to the one described above (except in Italian, of course.) Results showed that less than 10% of verbs with auxiliary essere &quot;to be&quot; pass the test for stativity. We also took the translations of the core set of \[+active\] and \[+stative\] verbs from the sample verbs given in Joos(1968) and in Quirk and Greenbaum(1972). We then examined the corresponding Italian verbs taken from the Collins English-Italian dictionary to determine which took essere &quot;to be&quot; and avere &quot;to have&quot;. Results showed that almost all of them took the auxiliary avere &quot;to have&quot;. Thus, we were unable to correlate activity and stativity with translation of verbs taking essere &quot;to be&quot; and avere &quot;to have&quot;. Conclusion. The active-stative distinction appears to be an aspectual feature related to discourse. It is not a lexical feature, although certain verbs occur more naturally whereas others are more tightly constrained. This means that the properties active and stative are qualitatively different from the inherent features \[+human\] or \[+male\] for nouns, or from the selectional features \[+thatcomp\] (takes a that complementizer), or \[requires a human subject\] for verbs. For example, a verb which indicates a true state, such as &quot;resemble&quot;, is more likely to occur in a sentence where stative aspect is present, but not necessarily. This being the case, then it is strictly speaking not the job of the lexicographer alone to identify such features. Rather, it is the job of the syntactician and semanticist to identify such contexts, along with the lexicographer who should identify verbs that are potentially active or stative in a given context.</Paragraph> <Paragraph position="52"> 4.2. SYNONYMY AND THE PROBLEM OF SENSE DISCRIMINATION.</Paragraph> <Paragraph position="53"> As described above, sense mapping between dictionaries is one of the most important and challenging problems facing computational lexicography. A related problem is sense disambiguation within a single dictionary. Webster's Seventh Collegiate Dictionary (W7) defines car as a kind of 'vehicle' but does not indicate that it is sense 4 of vehicle which is intended and not senses 1, 2, or 3. This lack of an explicit indication of the hypernym's intended sense limits the usefulness of the hypernym relation as we have currently extracted it because hypernymy is actually a relation between word senses, not a relation between words. As a consequence, our sprouts from hypernyms require human intervention to insure that the proper sense of each node is followed. Without automatic sense disambiguation, taxonomic sprouting is thus relegated to the status of a semi-automatic (rather than a fully automatic) processing tool.</Paragraph> <Paragraph position="54"> It has been suggested (Lesk(1986)) that sense disambiguation of hypernyms might be achieved by comparing the differentia (the words which appear with the hypernym, e.g., &quot;moving on wheels&quot; in the definition of car as a &quot;vehicle moving on wheels&quot;) to the words that appear in the definitions of the hypernym's senses. The sense definition which most closely matches the differentia would then be selected as the intended sense of the hypernym. The measure of closeness might be something as simple as the number of content words found in the intersection of words from the differentia and the sense definition. However, to the best of our knowledge, this has not yet been carried out automatically on a large scale. Our disambiguation of synonym senses has followed a somewhat different course based on the assumption that synonymy is a symmetric relation between word senses.</Paragraph> <Paragraph position="55"> Our procedures for sense disambiguation have been used to process The Collins Thesaurus (CT), which is stored as a DAM file with 16,700 keyed records containing a total of 278,000 synonym tokens. Each key is a headword or a boldface run-on word from a main entry. The record associated with each key holds the synonym tokens organized by part-of-speech of the key and, within each part-of-speech, by sense number.</Paragraph> <Paragraph position="56"> Many of the entries on the original tape and the printed version of CT are not marked for part-of-speech, so it was necessary to use UDICT to analyze the parts-of-speech of the synonyms listed under each sense number. From this it could then be inferred under which part-of-speech the sense number belonged (Chodorow and Ravin (1987)).</Paragraph> <Paragraph position="57"> In a dictionary-style thesaurus such as CT, an entry A may have word B listed as a synonym of its nth sense, and entry B may have word A listed as a synonym of its mth sense. Based on the assumption that synonyms are symmetric, we can mark B in entry A as the mth sense 234 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods of B, and A in entry B as the nth sense of A. An example of this type of one-to-one mapping in CT is given below. dense (adj) 1 .... condensed . . . solid ....</Paragraph> <Paragraph position="58"> 2 .... dull ...stupid ...</Paragraph> <Paragraph position="59"> dull (adj) 1. dense .... stupid ....</Paragraph> <Paragraph position="60"> 2 .... callous ... unsympathetic ....</Paragraph> <Paragraph position="61"> 7. drab . . . muted ....</Paragraph> <Paragraph position="62"> Here, sense 1 of dull is synonymous with sense 2 of dense. 37% of the 278,000 synonym tokens show this type of symmetry. Of course, there are also mappings of the one-to-many variety, but they account for only .5% of the tokens. We have automatically marked the senses of all synonyms in one-to-one and one-to-many relations. null The third type of mapping, many-to-many, accounts for just .5% of the total, but it poses a problem for the strategy outlined above. This can best be seen by considering an example. Senses 1 and 2 of institution list establishment as a synonym, and senses 1 and 2 of establishment list institution. Is sense 1 of institution synonymous with sense 1 of establishment or with sense 2? The distribution of the terms institution and establishment alone cannot answer the question, but there may an automatic solution based on the fact that synonyms tend to cluster into groups that represent separate senses. (This phenomenon will be described later in this section.) Consider again the case of dense and dull. Evidence for linking sense 2 of dense with sense 1 of dull comes from the symmetric distribution of the two words in the entries. There is however another piece of evidence for linking sense 2 of dense with sense 1 of dull, and that is the co-occurrence of the word stupid in both synonym lists. Thus, taking the intersections of synonym lists might form the basis for an automatic approach to disambiguating the many-to-many mappings. This is similar in some ways to Lesk's suggestion for disambiguating hypernyms by comparing words that appear with the hypernym to words that appear in the hypernym's sense definitions.</Paragraph> <Paragraph position="63"> Of course, not all tokens are symmetrically mapped; 62% appear in asymmetric relations. There appear to be many reasons for the asymmetry. 20% of the items offered as synonyms are phrases or rare words that simply do not appear as main entries in the thesaurus.</Paragraph> <Paragraph position="64"> The other 42% which do appear as main entries but do not 'point back' are the focus of current analysis. Some of the asymmetries appear to be mere oversights on the part of the lexicographers. For example, assembly has throng listed as a synonym of one of its senses, but throng does not list assembly as a synonym, although it does give assemblage, congregation, multitude, and other related words. A second type of asymmetry results when a central sense of one word is synonymous with a very peripheral sense of another. One sense of say lists add, as in &quot;He added that he would like to see the demonstration.&quot; The entry for add does not however contain this peripheral sense and deals only with the arithmetic add. This omission is more serious than simply omitting a word, for here an entire sense seems to have been left out. A third situation that gives rise to asymmetric links is when the relation between the two words is not really synonymy at all but rather hypernymy. For example, book lists manual as a synonym, but manual does not list book; instead special types of books such as handbook are given.</Paragraph> <Paragraph position="65"> Once asymmetries have been discovered, it is possible to supply main entries where they are missing (as in the case of rare words or phrases), to insert synonyms into already existing entries, and even to create missing senses of entries. These enhancements to the thesaurus should make it more valuable for computational research and also as a resource for human use.</Paragraph> <Paragraph position="66"> Other enhancements require a fundamental reorganization of information about synonyms. In a dictionary-style thesaurus, synonymy is treated as a relation between pairs of word senses. Another approach is to consider it a relation among words that share a common sense. It can thus be viewed as a many-to-one mapping between various spellings and a single meaning. The words dense, dull, and stupid all point to one another because they map to the same sense, although CT, as a dictionary-style thesaurus, has no direct way of indicating this fact.</Paragraph> <Paragraph position="67"> The many-to-one structure can be seen more explicitly in the traditional Roget's style of organization where a plan of classification clusters words into &quot;ideas&quot; and an alphabetical index of words is used to point to the ideas. In some dictionaries, such as W7, synonym paragraphs list synonyms or nearly synonymous words and describe the subtle, often connotative, differences between them. Main entries for the words point to the paragraphs in much the same way that alphabetically indexed words point to ideas in the traditional thesaurus. null Warnesson (1985) describes how words in a dictionary-style thesaurus can be automatically organized into synonym clusters, each representing a word-sense, and how associations between clusters can be measured. The goal of restructuring the thesaurus in this way is not to reproduce an already existing, externally imposed organizational scheme, such as Roget's, but rather to uncover the implicit structure of the thesaurus based on the actual interconnections among its words.</Paragraph> <Paragraph position="68"> Although we have not yet completed this work, we have developed the tools required for its first three stages. The first stage in the process is the disambiguation of senses, described above. Next, a set of related words is obtained by sprouting from a selected root node in the synonym index. Because the synonym senses have been disambiguated, no human intervention is required in the sprouting. The third stage is the production of a square matrix in which the rows and columns are labelled with the words obtained from sprouting, and Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 235 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods the cells contain either a &quot;1&quot; or a &quot;0&quot; to indicate the presence or absence of a synonym relationship between the row word and the column word. Finally, an algorithm (such as the one reported in Marcotorchino(1986)) is applied to the matrix to produce the clusters.</Paragraph> <Paragraph position="69"> There are several advantages to organizing synonyms into clusters, as suggested by Warnesson (1985). (1) Space can be saved by having all members of the cluster point to a single location which represents the sense they share, rather than listing in the entry for each member all the other members. (2) The word having the largest number of connections within the cluster and the fewest connections outside it might be offered to the user as the synonym which is most representative of the sense. (3) It is possible to compute the strength of association between clusters based on connections between member and nonmember words. Strong associations should exist between closely related senses, and weak associations between more distant ones. In this way, it might be possible to generate a hierarchical structure of senses from the bottom up, i.e., based on the actual patterns of synonym connections found in the thesaurus. (4) An on-line thesaurus in which sense-disambiguated words have been structured into clusters provides the best of both forms of organization.</Paragraph> </Section> <Section position="6" start_page="0" end_page="0" type="metho"> <SectionTitle> 5. APPLIED COMPUTATIONAL LEXICOLOGY: BILINGUAL DICTIONARIES. 5.1. TRANSFER OF LEXICAL INFORMATION THROUGH BILINGUAL DICTIONARIES. </SectionTitle> <Paragraph position="0"> The goal of the experiments described in this section is to take information available from a monolingual dictionary for a given language and to use a bilingual dictionary as a transfer device to enable the same information to be transferred to a monolingual dictionary in a second language. The clearest and most successful attempt at this is reported in the first section below on the semantic feature \[+human\]. Other more speculative attempts with synonyms and with the syntactic feature \[+thatcomp\] resulted in partial successes and are reported in later sections.</Paragraph> <Paragraph position="1"> 5.1.1. \[+HUMAN\] NOUNS.</Paragraph> <Paragraph position="2"> Using sprouting and filtering, we extracted a set of English nouns with the feature \[+Human\] in all senses.</Paragraph> <Paragraph position="3"> Next, we obtained the set of Italian translations for those nouns from the Collins English-Italian dictionary.</Paragraph> <Paragraph position="4"> Finally, we re-translated the Italian nouns back into English, using the Collins Italian-English dictionary.</Paragraph> <Paragraph position="5"> Our results were very positive. Of the 321 original English words, 296 Italian words were obtained in the forward (E-I) direction; 63 multi-word translations were also found. Re-translating these 296 Italian words in the backward (I-E) direction yielded 373 English words plus 45 English multi-word translations. Examining the two lists of English words showed that out of the 321 original, 157 were not among the final re-translations.</Paragraph> <Paragraph position="6"> Furthermore, out of the 373 re-translations, 167 were new words not in the original.</Paragraph> <Paragraph position="7"> An analysis of the words obtained gave two clear results. First, only six out of the 296 Italian words obtained in the forward (E-I) direction did not have the feature \[+ Human\] in all of their senses. This feature can therefore be very well transferred (given our constraint on the polysemy of the source data) to the other monolingual dictionary. Second, well over 90% of the 167 new English re-translations were also \[+Human\].</Paragraph> <Paragraph position="8"> This demonstrates the effectiveness of this procedure in the task of enlarging the set of words having a given feature.</Paragraph> <Paragraph position="9"> The procedures used for transferring the \[+human\] feature from English to Italian can be viewed as instances of a special kind of sprouting called &quot;bilingual sprouting&quot;. In this case, the binary relation through which we propagate features is the translation relation, rather than synonymy, hypernymy, or hyponymy. As with other forms of sprouting, we can view the translation relation as transitive, and therefore we are able to propagate lexical information not only from one language to another, but also back to the original language. Of course, again as with normal sprouting, we must take care to control for the effects of polysemy. These considerations lead to the definition of two types of bilingual sprouting: unconstrained and constrained.</Paragraph> <Paragraph position="10"> Unconstrained bilingual sprouting. This form of bilingual sprouting assumes that if a word has a particular feature in all of its senses, then any translations of that word that are monosemous will also bear that feature.</Paragraph> <Paragraph position="11"> This, of course, excepts grammatical features which only apply in one of the two languages involved. So, for example, we could not propagate grammatical gender from Italian to English.</Paragraph> <Paragraph position="12"> A reasonable procedure that implements this kind of sprouting would begin with a list of words in the first language (LI) bearing the feature to be propagated. For each word, assign the feature to any of its monosemous translations, given by the bilingual dictionary. This accomplishes the transfer of the feature to the second language (L2). We can continue by inspecting the bilingual dictionary for the other direction (from L2 to L1). Since the words in L2 are all monosemous, any translations from them back into L1 are also likely candidates for the feature.</Paragraph> <Paragraph position="13"> Despite the care taken by this procedure to deal only with monosemous words in L2, polysemy in L1 and in the bilingual dictionary causes the procedure to yield lists which must be manually checked. However, the labor involved is much less than would be required to do the work entirely by hand. As reported above, when the procedure was used to propagate the \[+Human\] feature from English to Italian, we obtained a list of Italian words which were correctly marked in much more than 95 percent of the cases. Furthermore, when we reflected the feature back onto English (with the Italian-English dictionary) we obtained a new set of English words not on our original list. More than 90 percent of those were indeed \[+Human\] words.</Paragraph> <Paragraph position="14"> 236 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods Constrained bilingual sprouting. This type of sprouting allows the assignment of features to word senses in L1, rather than to an entire word. It is based on the observation that if a single sense of a word in L1 has a monosemous translation in L2, then both words can be assigned any transferable features that either of them bears. This assumes that the lexicographers who created the dictionaries being used had comparable ideas about what constitutes a word sense. If we can maintain this assumption, then a sprouting procedure based on our observation should run automatically and yield quite reliable results. However, we still imagine that a manual check of the results for validity would be in order.</Paragraph> <Paragraph position="15"> A further observation is that if the word sense has multiple monosemous translations into L2, then each of the L2 words can receive the union of their transferable features. A possible algorithm based on this observation would use three dictionaries. First, a monolingual dictionary for L1 would contain features to be transferred. Second, a bilingual dictionary from L1 to L2 would serve as the conduit for the lexical information and the pointer to words in L2. Third, a monolingual dictionary for L2 would be the target for the information to be propagated. The algorithm would scan the words on the L1 side of the bilingual dictionary. For any word sense (in the monolingual dictionary for L1) with multiple monosemous translations into L2, we can assign the union of the transferable features (from the L1 word sense and all L2 translations) to to all of the words in the L2 monolingual dictionary.</Paragraph> <Paragraph position="16"> The two flavors of constrained bilingual sprouting (one for transferring information from L1 to L2, and the other for reflecting it back to L2) would both be strengthened if we could drop the requirement that the words in L2 be monosemous, and map to word senses instead. In the section 5.2.1, we discuss an enhancement to bilingual dictionaries which will have precisely this effect.</Paragraph> <Paragraph position="17"> Two important points must be made about bilingual sprouting. First, it is only feasible and useful when the lexical information available and desired in the two languages' monolingual dictionaries are the same. As reported above, for Italian and English we have established lists of features and attributes of interest for both languages, and we are confident that these lists match the requirements of prospective users of the monolingual dictionaries. Second, constrained bilingual sprouting obviously requires the prior establishment of the binary mapping relationships between both monolingual dictionaries and the respective sides of the bilingual dictionary.</Paragraph> <Paragraph position="18"> 5.1.2. SYNONYMS OF &quot;SAY&quot;.</Paragraph> <Paragraph position="19"> We have also made a preliminary attempt to transfer a thesaurus entry from the Collins Thesaurus (CT) into Italian by means of the English-Italian and Italian-English bilingual dictionaries. The goal of this effort is to determine the feasibility of using a monolingual English thesaurus and bilingual dictionaries to generate a monolingual Italian thesaurus. The entry selected for the test was the verb say, which has six senses listed in CT.</Paragraph> <Paragraph position="20"> First, the synonyms of each sense of say and their synonyms were obtained through sprouting to form a synonym set. Next, each of these synonyms was looked up in the English-Italian dictionary to obtain Italian translations. Of course, many of the translations were inappropriate. As an example, one translation of add is calcolare (&quot;to calculate&quot;). All of the Italian translations were then looked up in the Italian-English dictionary to obtain their English translations. These were compared to the original set of English synonyms. An Italian word was deleted if none of its English translations was among those in the original synonym set, except of course for the translation that was responsible for its being included in the first place. Calcolare was deleted because none of its English translations was found in the original group of English synonyms. In this way, inappropriate Italian words were eliminated from the translation set.</Paragraph> <Paragraph position="21"> A rectangular matrix was constructed with the columns representing English words and the rows representing Italian words. A &quot;1&quot; was placed at the intersection of a row and a column if the English word and the Italian word were listed as translations in the bilingual dictionary; otherwise the intersection held a &quot;0&quot;. An algorithm for clustering binary rectangular matrices (Marcotorchino(1986)) was used to rearrange the rows and columns.</Paragraph> <Paragraph position="22"> We were particularly interested in seeing if the clusters showed the structure of the original CT entry for say with its six senses. The results were somewhat mixed. For very circumscribed senses, such as the sense of say related to a performance (deliver, orate, perform; recite, etc.), the translations did indeed show the appropriate clustering, but for the more general senses, the structure, if any, was much harder to discern. Although this technique seems to hold some promise, it is too early to say if it is feasible to use an English thesaurus and bilingual dictionaries to automatically produce an Italian thesaurus.</Paragraph> <Paragraph position="23"> 5.1.3. \[+THATCOMP\] VERBS.</Paragraph> <Paragraph position="24"> We started with a list of English verbs which were hand-marked with a feature reflecting selection for a that complementizer. We constructed a matrix of grammaticality judgments for these verbs with the sentence frame generating program. The four test frame conditions were: (1) VERB obj that, (2) VERB ~b that, (3) VERB obj infinitival complement, and (4) VERB ~b.</Paragraph> <Paragraph position="25"> infinitival complement. We then took the translations of a test set of these verbs. The same syntactic tests were applied to the Italian verbs. An example of the combined matrix is given below. Each column refers to each syntactic test frame.</Paragraph> <Paragraph position="26"> Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 237 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods TEST 1 2 3 4 acknowledge ........... + + ammettere ........... + - + add ........................ + addizionare ..........</Paragraph> <Paragraph position="27"> admit ..................... + riconoscere .......... + - + - null The results we present should be viewed as preliminary because we took only a small set of some ten English verbs and some twenty-five Italian verbs. The first observation is that some of the translations do not belong to the same semantic field as the word in L1. An example is add and addizionare. Addizionare means to &quot;to add&quot; in the mathematical sense, but does not mean &quot;to add&quot; in the sense of &quot;comment&quot;. When we have lexical information applied to senses, disambiguated references in bilingual dictionaries (see section 5.2.1) and sense mappings (see section 5.2.2) such problems should disappear.</Paragraph> <Paragraph position="28"> Second, there are the cases of a single entry in L1 with multiple translations in L2. An example of that is anticipate: TEST l 2 3 4 anticipate ............... + aspettarsi ............ + - + precedere ............</Paragraph> <Paragraph position="29"> pregustare ...........</Paragraph> <Paragraph position="30"> prevedere ............ +- + prevenire ............. - + There is a possibility of having a that clause for at least one of the translations in each group of translations.</Paragraph> <Paragraph position="31"> Further, it appears that the pattern could serve as a clue to pick which of the many translations is the one that corresponds semantically to the English word sense taking that. This applies to the first two test frames, i.e. columns one and two in the table. For example in the case of anticipate, aspettarsi and prevedere are the two translations with the meaning closest to anticipate when anticipate is used in the sense taking a that clause. Our conclusion is that this process is not feasible completely automatically, but that such a procedure could reduce much searching and guessing. Further, we believe we could profit by utilizing information on that clauses found in example sentences in the bilingual dictionary. This could provide an additional mechanism to eliminate non-corresponding translations*</Paragraph> </Section> <Section position="7" start_page="0" end_page="0" type="metho"> <SectionTitle> 5.2. CROSS-REFERENCES AND SENSE DISAMBIGUATION IN BILINGUAL DICTIONARIES. </SectionTitle> <Paragraph position="0"> Since the Collins bilingual dictionaries play such an important role in our research, both as a conduit for lexical information transfer and as reference material for researchers, it is desirable to make them as accurate and useful as possible. This section discusses two methods of enhancing the contents of those bilingual dictionaries. The first method supplies sense numbers to the cross-references between the two sides of the dictionary. The second supplies &quot;missing information&quot; in the form of cross-references which should be there but are not.</Paragraph> </Section> <Section position="8" start_page="0" end_page="0" type="metho"> <SectionTitle> 5.2.1. DISAMBIGUATING REFERENCES IN COLLINS BILINGUAL DICTIONARIES. </SectionTitle> <Paragraph position="0"> The English-Italian side of Collins assigns sense numbers to the English words. Similarly, the Italian-English side assigns sense numbers to the Italian words. A desirable enhancement to the entries in either side is to assign sense numbers to the translations so that they accurately point to the other side. Pictorially, we want to instantiate sense references represented by the arrows in Figure 9. This result is essentially equivalent to that discussed in section 4.2 for disambiguating the cross-references in the Collins Thesaurus (CT).</Paragraph> <Paragraph position="1"> In terms of LDBs, we are proposing that the structure of the current translation attribute &quot;xlat&quot; be enhanced by the addition of a new &quot;senseno&quot; (sub)attribute identifying the sense number on the other side of the dictionary for the word given as the value of the &quot;spel&quot; (sub)attribute. The resulting &quot;xlat&quot; attribute will look like:</Paragraph> <Paragraph position="3"> As with the mapping files, several procedures will be required in order to determine the correct values for these new &quot;senseno&quot; attributes. Some of these procedures will be trivial, as when the words involved are monosemous. Recall that, in the case of sense mapping with Webster's Seventh, the majority of words are monosemous. We expect to find the same situation in Collins. A version of the procedure given in Section 4.2 should supply many values. Specifically, if any sense of an English word, El, gives a particular Italian word, I1, as its translation, and if one sense of I1 gives E1 as its translation, then we can mark the II translation of El (on the English-Italian side) with the sense number of I 1 (on the Italian-English side) that had the reciprocal 238 Computational Linguistics, Volume 13, Numbers 3-4, July-December 1987 Roy J. Byrd, Nicoletta Calzolari, Martin S. Chodorow, Judith L. Klavans, Mary S. Neff, Omneya A. Rizk Tools and Methods reference. Further, more difficult, and possibly even manual procedures may be required to obtain the rest.</Paragraph> <Paragraph position="4"> In any event, once the sense links have been established, they will remain permanently valid, as is the case for the inter-dictionary mapping files.</Paragraph> <Paragraph position="5"> An immediate benefit that would emerge from disambiguating the references in a bilingual dictionary would be that we can increase the power of constrained bilingual sprouting. We would then be able to map directly onto word senses of polysemous words in the second language and would not be restricted to using only monosemous words. The reader is invited to refer back to the discussion of bilingual sprouting, in section 5.1.1, for details.</Paragraph> </Section> <Section position="9" start_page="0" end_page="0" type="metho"> <SectionTitle> 5.2.2. SYMMETRIZING COLLINS BILINGUAL DICTIONARIES. </SectionTitle> <Paragraph position="0"> Inspection of the Collins English-Italian and Italian-English dictionaries reveals that there are many asymmetries between the two sides. An example of an asymmetry would be a case where an English word gives an Italian word as its translation, but that Italian word does not, in turn, give the original English word as its translation. In order to provide improved access to bilingual information by humans and by our analysis procedures, it seems feasible to produce a symmetrized bilingual dictionary. Essentially, this would involve automatically locating all asymmetries and adding additional entries and/or translations to make them symmetric. It should not be difficult to accomplish this task, given LDBs for the two dictionaries.</Paragraph> <Paragraph position="1"> Despite its initial appeal, we suspect that the creation of a fully symmetrized bilingual dictionary may not be totally desirable. It may be that some of the asymmetries exist because of valid lexicographic principles, and that violation of those principles will lead to an undesirable result. Through inspection of the dictionary we have pinpointed four candidate principles which could warrant the use of asymmetric references.</Paragraph> <Paragraph position="2"> 1. Lemma forms of words which are only (or mostly) used in derived forms should not be given as translations of words in the other language. Thus, although allege appears in the English-Italian dictionary, it is never given as the translation of an Italian word. This may be because allege is almost always used in English in its past participial form: alleged. In the corpus analyzed in Kucera and Francis (1967), alleged occurs ten times while allege occurs only once.</Paragraph> <Paragraph position="3"> 2. Rare words or words referring to highly specific concepts may be translated with more general words, however a general word should not be translated with a specific or rare one. Thus, English moment is translated as Italian importanza in its sense of &quot;importance&quot;. However, since the &quot;importance&quot; meaning of English moment is quite specific, it is not given as the translation of Italian importanza. This case seems similar to the example of say and add given in the discussion of asymmetry in CT. In the case of bilingual dictionaries, perhaps the principle is to avoid offering such rare word senses to the non-native speaker, who would find them difficult to use correctly.</Paragraph> <Paragraph position="4"> 3. In apparent violation of the preceding principle, we sometimes find a general word translated with what are essentially hyponyms in the other language. Thus, for book, Collins English-Italian gives quaderno &quot;(notebook)&quot;, bustina &quot;(of matches)&quot;, and blocchetto &quot;(of tickets)&quot;. The Italian-English side of Collins only gives specific translations for these words; none is translated as book. This seems entirely equivalent to the case of book and manual, covered in the discussion of CT. The principle here may be to avoid giving as translations either hypernyms or generic words which acquire a particular meaning only in a specific context. In the case of &quot;book&quot;, the hypernym would be completely redundant with respect to the specific translations that are given and would thus add no new information.</Paragraph> <Paragraph position="5"> 4. Some translations are not symmetric because the translation given does not appear as an entry on the other side of the dictionary. For example, Collins lists to parse as one of the English translations for Italian analizzare. However the English-Italian side of Collins does not contain the word parse. Similarly, for Italian codifica, Collins gives codification and for Italian proclamare it gives promulgate; neither of these English words appears as a head word on the English-Italian side of the dictionary. These omissions may merely reflect an oversight on the part of the lexicographers. null We assume that principles whose only purpose is to save space can be safely violated by the creation of the missing symmetric links. Our approach to sorting out the question of whether symmetrizing is a good thing in general is to generate candidate links and to see if they violate these or other principles and whether we like the result. We offer the preceding list of types of asymmetry not as hypotheses to be verified, but merely as things to look for during such an analysis. In the process, we may discover the real principles which favor the creation of asymmetric links, if any exist. The best outcome may be that we can build a partially symmetrized dictionary containing all desirable symmetry and no undesirable asymmetry. A further benefit is that we can provide a tool which allows the lexicographers to inspect the asymmetries in their dictionaries before committing them to print.</Paragraph> </Section> class="xml-element"></Paper>