ࡱ> G ^bjbjَ IW]  x x x $X X X X P 4 ,X +L F\0/J$wkfx ,,,,x x  ,,,,x x  @ | @8 @x x ,,h,x x  ߟQVX X , Ambiguity in the Acquisition of Lexical Information Lucy Vanderwende Microsoft Corporation One Microsoft Way Redmond, WA 98052-6399 lucyv@microsoft.com Abstract This paper describes an approach to the automatic identification of lexical information in on-line dictionaries. This approach uses bootstrapping techniques, specifically so that ambiguity in the dictionary text can be treated properly. This approach consists of processing an on-line dictionary multiple times, each time refining the lexical information previously acquired and identifying new lexical information. The strength of this approach is that lexical information can be acquired from definitions which are syntactically ambiguous, given that information acquired during the first pass can be used to improve the syntactic analysis of definitions in subsequent passes. In the context of a lexical knowledge base, the types of lexical information that need to be represented cannot be viewed as a fixed set, but rather as a set that will change given the resources of the lexical knowledge base and the requirements of analysis systems which access it. Introduction In order to identify lexical information automatically in an on-line dictionary, various techniques have been used, all of which employ defining formulae in some way. Defining formulae are significant recurring phrases in dictionary definitions (Markowitz et al. 1986; p. 113). Some have implemented these defining formulae as string patterns that match the definition text, e.g., Chodorow et al. (1985) and Markowitz et al. (1986), while others have implemented them as structural patterns that match the syntactic analysis of the definition text, e.g., Jensen and Binot (1987), Alshawi (1989), Ravin (1990), Montemagni and Vanderwende (1992), and Klavans et al. (1990). The earlier studies used defining formulae as if there were a one-to-one relation between the formula and the type of lexical information it identified; for example, the defining formulae used for always identifies the Instrument type of lexical information. Later studies (Ravin 1990 and Klavans et al. 1990) have shown that some defining formulae can convey several types of lexical information. The problem is that defining formulae sometimes must rely on ambiguous words in their patterns, such as with, of and unit of as shown below, and even in dictionaries which make use of a limited defining vocabulary, this ambiguity is not resolved. Ravin (1990) shows that with in the definitions of verbs can convey many types of semantic relations, e.g., the relation USE-OF-INSTRUMENT in the definition of angle (L 3,vi,1): to fish with a hook and line, and MANNER in the definition of attack (L 1,v,4): to begin (something) with eagerness and great interest. In Klavans et al. (1990), we see that the pattern a unit of can convey the relation AMOUNT/CURRENCY in the definition of pice (W n,1): a former monetary unit of India and Pakistan equal to ..., and it can convey SUBDIVISION in the definition of division (W n,3c1): the basic unit of men for administration .... Both of these studies argue that a syntactic analysis of the dictionary definition is required in order to ide@ify the lexical information reliably. Klavans et al. (1990) show how the relation conveyed by the pattern a unit of can be identified on the basis of syntactic information. Ravin (1990) describes a system for disambiguating the preposition with in verb definitions which requires a syntactic analysis of the definition, but also lexical information for the verb modified by with and the noun complement of with. For example, to determine that with conveys a USE-OF-INSTRUMENT relation in the definition of angle: to fish with a hook ..., lexical information is needed for both the verb fish and the noun hook. In order to acquire this lexical information, the definitions of fish and hook are parsed, after which heuristics, or patterns, are applied to the syntactic analysis to determine the necessary lexical information. The process described in Ravin (1990) can be seen as depth-first: in order to acquire the semantic relations for the definition of angle, at least some of the semantic relations for other words needs to be acquired first. In this paper, I will describe a different approach to solving this problem from that taken in Ravin (1990). This implementation involves multiple passes through the dictionary, and at each stage all the lexical information that can be reliably identified is added to the lexical knowledge base (LKB). We store the lexical information in the manner described in Dolan et al. (1993) and Richardson et al. (1993). This process can be seen as breadth-first: acquire lexical information for each word in the dictionary, then use that information to acquire more and/or more reliable lexical information. When all we are concerned with is to disambiguate the defining formula, the results of our approach will not differ from those of Ravin. However, when acquiring lexical information from definitions which are syntactically ambiguous, our approach shows better results. Processing the dictionary multiple times During the first pass of our incremental approach, only the defining formulae which unambiguously identify lexical information will be used. Based on the syntactic analysis of a broad-coverage parser, such as PEG (Jensen 1986) used in Ravins study and the Microsoft English Grammar used in this study, a number of semantic relations can be identified, e.g., HYPERNYM, INSTRUMENT-OF, MATERIAL, PART, and PART-OF. For some of these relations, not all of their possible patterns, or defining formulae, will be applied at this stage, e.g., for PART-OF, the unambiguous pattern part of can be applied, but not the pattern of, which will be discussed later. The lexical information in (= seed1) has been identified by applying the part of pattern to the syntactic analysis of the definition of flower (L 1,n,1): the part of a plant, often beautiful and colored, ... (seq exs1) [flower] SYMBOL 174 \f "Symbol" (PART-OF) SYMBOL 174 \f "Symbol" [plant] Similarly, for the relation PART, the unambiguous pattern {that,which} {has,have} can be applied; the lexical information in (= plant2) has been identified by applying this pattern to the syntactic analysis of the definition of plant (L 2,n,1): a living thing that has leaves and roots, and grows usu. in earth, ... . (seq exs2) [plant] SYMBOL 174 \f "Symbol" (PART) SYMBOL 174 \f "Symbol" [leaf, root] Once the semantic relations acquired during the first pass have been added to the LKB, the lexical information exists which will allow the ambiguous words in the defining formulae to be disambiguated, enabling the identification of more semantic relations during subsequent passes. One of the patterns for identifying a PART-OF relation which could not be applied during the first pass is the pattern characterized by of. The preposition of can convey many different relations, among which are PART-OF, MATERIAL, and HYPERNYM. Examples of definitions in which of conveys PART-OF, MATERIAL, and HYPERNYM relations are (= bulb3)-(= christening5), respectively. (seq exs3) clove (L 1,n): the dried unopened flower of a tropical Asian plant, used ... (seq exs4) bullion (L n): bars of gold or silver (seq exs5) christening (L,n): the Christian ceremony of baptism ... In order to determine the relation conveyed by of, it is necessary to access the lexical information of the noun modified by of and the complement of of. The pattern which identifies PART-OF from an of prepositional phrase (of-PP) can be paraphrased as: (a) if the modified noun has a PART-OF relation whose value matches the of-complement, or (b) if the complement has a PART relation whose value matches the modified noun, then the of-PP conveys the relation PART-OF. Consider the definition of clove (L 1,n): the dried unopened flower of a tropical Asian plant, used ... Condition (a) applies to this definition, because the modified noun, flower, has a PART-OF relation, as shown in (= seed1), and its value, namely plant, is the same as the complement of the of-PP. The lexical information identified from this definition is shown in (= clove6). (seq exs6) [clove] SYMBOL 174 \f "Symbol" (PART-OF) SYMBOL 174 \f "Symbol" [plant] The results of the incremental approach described until now do not differ from those of Ravin; both can handle ambiguous defining formulae well. The only difference is that in our approach, the lexical information for flower and plant will already be available when the entry for clove is being processed, while in Ravins approach the definitions of flower and plant will need to be analyzed before the PART-OF relation can be identified in the definition of clove. The incremental approach shows clear advantage, however, in acquiring lexical information from definitions which are syntactically ambiguous. The semantic relations that have been acquired on a first pass not only serve to disambiguate defining formulae, but they can also be used to improve the syntactic analysis of dictionary definitions in subsequent passes. Syntactic analysis during subsequent passes The syntactic analysis to which the patterns are applied is the output of the Microsoft English Grammar, a broad-coverage grammar. Guided by rule and part-of-speech probabilities, the parsing algorithm produces the most probable parse first (see Richardson, 1994). The method for dealing with ambiguity at this initial, syntax-only, stage is to attach pre- and post-modifiers in a single arbitrary pattern (usually to the closest possible head, ...) (Jensen and Binot 1987). In order to solve the ambiguities of the initial syntactic analysis, Jensen and Binot (1987) proposed to access the semantic information that can be identified automatically in an on-line dictionary. The goal of the work under discussion is to provide this level of semantic information, and the relations that can be identified in a definition during the first pass of the on-line dictionary fulfill this goal, but only when the definition is not syntactically ambiguous. Most dictionary definitions, however, are syntactially ambiguous. All of the sources of ambiguity that occur in free text are also found in definition texts: prepositional phrases, relative clauses, and participle clauses can all be multiply attached. The most common source of ambiguity in definitions, however, is coordination. While the method of attaching post-modifiers to the closest possible head (i.e., right attachment) produces the correct parse in the vast majority of cases, this method is less successful in solving ambiguity in coordination. Consider the definition of plantain in (= plantain7): (seq exs7) plantain (L n): a type of common wild plant with wide leaves growing close to the ground and small green flowers The placement of the constituent small green flowers is ambiguous; small green flowers can be coordinated with ground to form the constituent the ground and small green flowers, or it can be coordinated with leaves to form the constituent wide leaves and small green flowers. Because the initial syntactic analysis arbitrarily attaches an ambiguous constituent to its closest head, small green flowers forms a constituent with ground. This is illustrated in the syntactic analysis in figure 1, which has been excerpted to focus on the coordination ambiguity. Based on this analysis, the pattern for the PART relation will identify that the plantain has leaves, but it will fail to find that the plantain has flowers. EMBED MSDraw \* mergeformat Figure 1. Schematic representation of the parse for the definition of plantain (L n) before reattachment One of the heuristics for handling coordination is to check which terms are most similar. In this example, we should check whether ground and flower are more similar than leaf and flower are. To check the similarity, the lexical information acquired during the first pass from the definitions of ground, leaf, and flower will be compared. As we saw in (= seed1) and (= plant2), repeated here as (= flower28) and (= plant29), both flower and leaf are parts of a plant. (seq exs8) flower (L 1,n,1): [flower] SYMBOL 174 \f "Symbol" (PART-OF) SYMBOL 174 \f "Symbol" [plant] (seq exs9) plant (L 2,n,1): [plant] SYMBOL 174 \f "Symbol" (PART) SYMBOL 174 \f "Symbol" [leaf, root] This similarity between leaf and flower ranks higher than the similarity found between ground and flower. The closest connection between ground and flower was found through the word grow, namely that flowers are grown (see flower (L 1,n,1) and that growing is located on the ground (see, e.g., gourd (L n,1)). Given the higher similarity between leaf and flower, the initial syntactic analysis is modified to reflect that wide leaves and small green flowers is a constituent; the revised analysis is shown in figure 2. Based on the revised syntactic analysis, the pattern for the PART relation will now identify that plantain has both leaves and flowers. EMBED MSDraw \* mergeformat Figure 2. Schematic representation of the parse for the definition of plantain (L n) after reattachment Patterns and heuristics are merging While the first pass identifies relations by using the initial syntactic analysis of the definitions, this example shows that more relations can be identified by disambiguating the initial syntactic sketch. And we saw that by disambiguating the defining formulae, more accurate relations can be identified. Sometimes disambiguating the initial syntactic sketch involves disambiguating (part of) the defining formulae, e.g., when determining the correct attachment of a PP. In this case, it is interesting to note that the patterns for identifying a particular relation are merging with, if not already quite the same as, the heuristics for determining the correct attachment. Consider the definition of angling (L n): the sport of catching fish with a hook and line. According to the strategy of right attachment, the initial syntactic analysis attaches the PP with a hook and line to the noun fish; the initial analysis is shown in figure 3. In order to determine the correct attachment of the PP, the system compares (a) to catch with a hook and line to (b) a fish with a hook and line (see Jensen and Binot (1987)). One of the reattachment heuristics checks for an INSTRUMENT relation between a head and the complement of with. During the initial pass, the INSTRUMENT relation in (= hook10) was identified from the definition of hook (L 1,n,1): a curved piece of metal, plastic, etc., for catching something ... . (seq exs10) [hook] SYMBOL 174 \f "Symbol" (INSTRUMENT) SYMBOL 174 \f "Symbol" [catch] Because there is an INSTRUMENT relation between the head catch and the complement hook, the analysis in (a) to catch with a hook and line is preferred, as shown in figure 4. EMBED MSDraw \* mergeformatFigure 3. Schematic representation of the parse for the definition of angling (L n) before reattachment EMBED MSDraw \* mergeformatFigure 4. Schematic representation of the parse for the definition of angling (L n) after reattachment One of the patterns that are applied to the syntactic analysis is the following INSTRUMENT pattern, which can be paraphrased as: if the with-PP modifies a verb, and if the complement of with has an INSTRUMENT relation whose value matches the modified verb, then identify an INSTRUMENT relation on the verb. We can see that the initial syntactic analysis of the definition of angling did not meet the conditions of this INSTRUMENT pattern because the with-PP modified the noun fish, and not the verb catch. The revised syntactic analysis, however, does meet the conditions and so the INSTRUMENT relation in (= angling11) was identified: (seq exs11) [angling] SYMBOL 174 \f "Symbol" (INSTRUMENT) SYMBOL 174 \f "Symbol" [hook, line] We can see that the heuristics for attaching a with-PP are quite similar to the pattern for identifying an INSTRUMENT relation from a with-PP. Both the task of determining the correct attachment and of determining the semantic relation need to access lexical information for the words in the text, whether free text or dictionary text. The endeavor of discovering defining formulae and formulating their patterns has sometimes been criticized as "open-ended", due to the variation with which relations are specified in the dictionary (Veronis and Ide, 1993). While it is certainly the case that there are many ways in which a PART-OF or INSTRUMENT relation is conveyed, these patterns must be seen as a reusable tool, one which can also help solve ambiguity in free text. Moreover, these patterns can identify semantic relations in free text, as well as in dictionary text, and so they should be seen as part of a more general system for language understanding. Conclusion The overall goal for acquiring lexical information automatically is to create an LKB that can be used in syntactic and semantic processing; in particular, lexical information that can motivate the correct analysis of PPs, relative clauses, and coordination. The task of processing dictionary definitions requires the correct analysis of the same phenomena just mentioned, as well as the disambiguation of defining formulae. By adopting an incremental approach to the acquisition of lexical information, we have seen that information extracted during the initial pass can be used to disambiguate the analyses and defining formulae during subsequent pass, enabling the acquisition of more, and more accurate, lexical information. The approach is, therefore, not circular, a concern put forward by Veronis and Ide (1993); at no time is the ambiguity in one definition solved by chasing through the definitions of other definitions, a situation which quickly could lead to an infinite loop. The apparent circularity is managed by storing the results of each pass in the LKB for use during a subsequent pass over the dictionary, where the LKB improves in quality after each new pass. An important result of implementing the patterns for identifying lexical information during subsequent passes is the discovery that they are similar to the heuristics for solving ambiguity problems in free text. In our case, we see these two sets of rules merging with no, or very few, differences. The dictionary patterns should be seen as a reusable tool and not of use only within the context of dictionary definitions. Finally, taking an incremental approach to the acquisition of lexical information raises an important issue in the context of the representation of lexical information. As a system develops, different types of lexical information will need to be represented; at first, only a subset of the possible relations will be available. As the paper has shown, we can get better results from processing the dictionary automatically if we do not expect all of the lexical information to be there at once. A flexible representation schema is therefore necessary. Also, NLP is not at a stage where it is known a priori which semantic relations are necessary; at best, we currently have some studies of the semantic relations which are minimally required by a specific component (e.g., Vanderwende (1994) lists the relations necessary for a noun sequence analysis). We must first explore which semantic relations are needed, before trying to fix upon the ultimate representation of lexical information. Acknowledgments I would like to thank Steve Richardson and George Heidorn for implementing the control structure which underlies the system described here; and Karen Jensen, Bill Dolan, Joseph Pentheroudakis and Diana Peterson for their feedback and support. References Alshawi, H. 1989. Analysing the Dictionary Definitions. In Boguraev & Briscoe, eds., Computational Lexicography for Natural Language Processing, Longman, London, pp. 153-170. Chodorow, M.S., R.J. Byrd, G.E. Heidorn. 1985. Extracting semantic hierarchies from a large on-line dictionary. In Proceedings of the 23rd Annual Meeting of the ACL, pp. 299-304. Dolan, W.B., L. Vanderwende and S.D. Richardson. 1993. Automatically deriving structured knowledge bases from on-line dictionaries. In Proceedings of the First Conference of the Pacific Association for Computational Linguistics, at Simon Fraser University, Vancouver, BC., pp.5-14. Jensen, K. 1986. Parsing strategies in a broad-coverage grammar of English. Research Report RC 12147, IBM Thomas J. Watson Research, Yorktown Heights, NY. Jensen, K., and J.-L. Binot. 1987. Disambiguating prepositional phrase attachments by using on-line dictionary definitions. In Computational Linguistics 13.3-4.pp. 251-60. Klavans, J., M. Chodorow, and N. Wacholder. 1990. From Dictionary to Knowledge Base via Taxonomy. In Electronic Text Research, University of Waterloo, Centre for the New OED and Text Research, Waterloo, Canada. Markowitz, J., T. Ahlswede and M. Evens. 1986. Semantically Significant Patterns in Dictionary Definitions. In Proceedings of the 24th Annual Meeting of the Association for Computational Linguistics, June 10-13 1986, pp. 112-119. Montemagni, S. and L. Vanderwende. 1992. Structural Patterns vs. string patterns for extracting semantic information from dictionaries. In Proceedings of COLING92, pp.546-552. Ravin, Y. 1990. Disambiguating and Interpreting Verb Definitions. In Proceedings of the 28th Annual ACL Conference, June 6-9. Pittsburgh, pp. 260-267. Richardson, S.D., L. Vanderwende and W.B. Dolan. 1993. Combining dictionary-based methods for natural language analysis. In Proceedings of TMI-93, Kyoto, Japan. pp.69-79. Richardson, S.D. 1994. Bootstrapping statistical processing into a rule-based natural language parser. In Proceedings of the ACL Workshop "Combining symbolic and statistical approaches to language", pp. 96-103. Vanderwende, L. 1994. Algorithm for Automatic Interpretation of Noun Sequences. In Proceedings of COLING94, pp.782-788. Veronis, J. and N. Ide. 1993. Extracting knowledge bases from machine-readable dictionaries: have we wasted our time? In Proceedings of KB&KS '93, pp.257-266. The sense numbers are proceeded by the letter L, which indicates that the source of the definition is Longman Dictionary of Contemporary English. There are further restrictions to the pattern part of, namely that part should be identified as the HYPERNYM. For the sake of brevity, the patterns will be given only in abbreviated form in this paper. For a more detailed account of structural patterns, see Montemagni and Vanderwende (1992). First, the relation [catch] SYMBOL 174 \f "Symbol" (INSTRUMENT) SYMBOL 174 \f "Symbol" [hook, line] is identified, but then also the relation in (= angling11) [angling] SYMBOL 174 \f "Symbol" (INSTRUMENT) SYMBOL 174 \f "Symbol" [hook, line], because catch, as well as sport, is the HYPERNYM of angling. 56Gmn{   " ' . JOz{%dh .2TX?CHLX]F#*bhE jCJU j0J$U6CJCJCJ55CJ5CJW56G]omn{& &F$d$d $dx$dx $$dx$dx$dz{BB[[[\\C]D]}^~^^^^^^F OH '!("""#$$&-(.(Z(,~..1 $$dx $ d $ d$d$dx $dxE[PQXYZ[fg}~OQef  ! " # & ' 4 5 6 7 I J Q R S T V [ V!X!!!!!!!" "p"r"""#$#########$ $2$4$$$ jCJUCJ6CJ_$$$$$$$$$$$$$$$$$$$%%%%&&I&O&T&Y&&&.(Z(`.h.m.n.x.y.z.{......... /3/B/U/n/t//////0~00001111111111111192A22222 3 333 j CJUjCJUVmH56CJ jCJUCJX111\2]2@44&577C8D8h8;>a>????"@#@BC $$dx $dx$dx$d%d&d'd$dx33333333333333333333333334444444 4%4)4D4E4L4M4N4O4R4X4m4n44444444444444444444455>5B5G5M5}55555555556 6M6R6666666777777 jCJU6CJCJ`7777777!8)8C8D8h8*;1;;;;;o<<<<:=>=v=w=}=~=====>> > > > >>>/>0>?>@>V>W>>>>>>>??.?/?0?1?w?~???????@@@@@@A jACJUj6CJUVmH jD1CJUj&CJUVmH55CJ6CJ j CJUjCJUVmH jCJUCJHAAAABBBBBBBBBBBBBBBBBBBBBBBBCCDCHCCCFFFQ%QR$R[[\\J\Q\_\c\C]D]`]a]w]x]]]]]]]]]]]]] ^ ^^^1^2^I^N^[^`^u^|^^^ jU65 j0J$U jCJUCJ6CJPChDFFFIK1MQQ%QRR$RRST   ."Arial- 2 M NP 2 NP 2 NOUN"Arialn- 2 CONJ 2 NOUN"Arial- 2 !leaves 2 Wflowers 2 FPRPRPTCL2 ONP"Arialn- 2 NOUN"Arial- 2 Wground- & MrEd% 2 c plant with & MrEd%F!! 2 =growing & MrEd%P` & MrEd%P` & MrEd%! & MrEd%F! 2 Wcand & MrEd%&1 & MrEd%1 & MrEd%_J & MrEd%J & MrEd%J & MrEd% l & MrEd%& 2  close to the "System--''erJet IIISi PostScriptpscriptLPT1:RSIZEDLRVAWLWRW YAYial METAFILEPICT(;((  Z&MrEdMicrosoft DrawZ&MrEd`  System)_  &MrEd h  d>   ."Arial- 2 M NP 2 NP 2 NOUN"Arialn- 2 CONJ 2 NOUN"Arial- 2 !leaves 2 Wflowers 2 FPRPRPTCL2 ONP"Arialn- 2 NOUN"Arial- 2 Wground- & MrEd% 2 c plant with & MrEd%F!! 2 =growing & MrEd%P` & MrEd%P` & MrEd%! & MrEd%F! 2 Wcand & MrEd%&1 & MrEd%1 & MrEd%_J & MrEd%J & MrEd%J & MrEd% l & MrEd%& 2  close to the "System--''erJet IIISi PostScriptpscriptLPT1:RSIZEDLdD(z  Z&MrEdMicrosoft DrawZ&MrEd`  System)_  &MrEd h  d>   ."Arial- 2 M NP 2 NP 2 NOUN"Arialn- 2 CONJ 2 NOUN"Arial- 2 !leaves 2 Wflowers 2 FPRPRPTCL2 ONP"Arialn- 2 NOUN"Arial- 2 Wground- & MrEd% 2 c plant with & MrEd%F!! 2 =growing & MrEd%P` & MrEd%P` & MrEd%! & MrEd%F! 2 Wcand & MrEd%&1 & MrEd%1 & MrEd%_J & MrEd%J & MrEd%J & MrEd% l & MrEd%& 2  close to the "System--''erJet IIISi PostScriptpscriptLPT1:RSIZEDL MSDraw`9  Z&MrEdMicrosoft DrawZ&MrEd k System)_  &MrEd    ."Arial- 2 M NP "ArialL-2 @NP"Arial- 2 NOUN"ArialL- 2 @CONJ"Arial- 2 NOUN 2 leaves 2 flowers 2 PRPRPTCL2 NP- & MrEd% 2 m plant with & MrEd%2 growing close to & MrEd%P` & MrEd%P` & MrEd% & MrEd% 2 and & MrEd%@@ & MrEd%@ & MrEd% & MrEd%` & MrEd%P & MrEd% 2 Q the ground"Arial-2 @NP & MrEd%CSC "System--''"Arial - METAFILEPICTH9   Z&MrEdMicrosoft DrawZ&MrEd k System)_  &MrEd    ."Arial- 2 M NP "ArialL-2 @NP"Arial- 2 NOUN"ArialL- 2 @CONJ"Arial- 2 NOUN 2 leaves 2 flowers 2 PRPRPTCL2 NP- & MrEd% 2 m plant with & MrEd%2 growing close to & MrEd%P` & MrEd%P` & MrEd% & MrEd% 2 and & MrEd%@@ & MrEd%@ & MrEd% & MrEd%` & MrEd%P & MrEd% 2 Q the ground"Arial-2 @NP & MrEd%CSC "System--''"Arial -D9kw M  Z&MrEdMicrosoft DrawZ&MrEd k System)_  &MrEd    ."Arial- 2 M NP "ArialL-2 @NP"Arial- 2 NOUN"ArialL- 2 @CONJ"Arial- 2 NOUN 2 leaves 2 flowers 2 PRPRPTCL2 NP- & MrEd% 2 m plant with & MrEd%2 growing close to & MrEd%P` & MrEd%P` & MrEd% & MrEd% 2 and & MrEd%@@ & MrEd%@ & MrEd% & MrEd%` & MrEd%P & MrEd% 2 Q the ground"Arial-2 @NP & MrEd%CSC "System--''"Arial - MSDraw@"C Z&MrEdMicrosoft DrawZ&MrEd Arial)_  &MrEd    ."Arialkv-2 `NP 2 PDET 2 PNOUN2 PPP 2 PREP 2 p VERB2  VP"Arial-2 pNP 2 NOUN2  PP"Arialkv- 2 the 2 sport2 pof 2 catching 2 fish 2  with a hook- & MrEd% & MrEd% & MrEd%0 & MrEd%0P0 & MrEd%0PP & MrEd%PP & MrEd%P & MrEd%p & MrEd%pP 2  and line "System--''$$$$$$$ METAFILEPICT"("Cng Z&MrEdMicrosoft DrawZ&MrEd Arial)_  &MrEd    ."Arialkv-2 `NP 2 PDET 2 PNOUN2 PPP 2 PREP 2 p VERB2  VP"Arial-2 pNP 2 NOUN2  PP"Arialkv- 2 the 2 sport2 pof 2 catching 2 fish 2  with a hook- & MrEd% & MrEd% & MrEd%0 & MrEd%0P0 & MrEd%0PP & MrEd%PP & MrEd%P & MrEd%p & MrEd%pP 2  and line "System--''dD"C  Z&MrEdMicrosoft DrawZ&MrEd Arial)_  &MrEd    ."Arialkv-2 `NP 2 PDET 2 PNOUN2 PPP 2 PREP 2 p VERB2  VP"Arial-2 pNP 2 NOUN2  PP"Arialkv- 2 the 2 sport2 pof 2 catching 2 fish 2  with a hook- & MrEd% & MrEd% & MrEd%0 & MrEd%0P0 & MrEd%0PP & MrEd%PP & MrEd%P & MrEd%p & MrEd%pP 2  and line "System--''4 MSDraw(" Z&MrEdMicrosoft DrawZ&MrEd Arial)_  &MrEdO    ."Arial-2 `NP 2 PDET 2 PNOUN2 PPP 2 PREP"Arial߂- 2  VERB2  VP"Arial- 2 pNOUN"Arial߂-2 PP"Arial- 2 the 2 sport2 pof 2 0catching 2 0pfish 2 0 with a hook- & MrEd% & MrEd% & MrEd%0 & MrEd%0P0 & MrEd%0PP & MrEd%P@P & MrEd%P@ & MrEd%P@ 2 and line "System--''$$$"C Z&* 4V: METAFILEPICT("eh("N Z&MrEdMicrosoft DrawZ&MrEd Arial)_  &MrEdO    ."Arial-2 `NP 2 PDET 2 PNOUN2 PPP 2 PREP"Arial߂- 2  VERB2  VP"Arial- 2 pNOUN"Arial߂-2 PP"Arial- 2 the 2 sport2 pof 2 0catching 2 0pfish 2 0 with a hook- & MrEd% & MrEd% & MrEd%0 & MrEd%0P0 & MrEd%0PP & MrEd%P@P & MrEd%P@ & MrEd%P@ 2 and line "System--''$$$"C Z&D("^ Z&MrEdMicrosoft DrawZ&MrEd Arial)_  &MrEdO    ."Arial-2 `NP 2 PDET 2 PNOUN2 PPP 2 PREP"Arial߂- 2  VERB2  VP"Arial- 2 pNOUN"Arial߂-2 PP"Arial- 2 the 2 sport2 pof 2 0catching 2 0pfish 2 0 with a hook- & MrEd% & MrEd% & MrEd%0 & MrEd%0P0 & MrEd%0PP & MrEd%P@P & MrEd%P@ & MrEd%P@ 2 and line "System--''$$$"C Z&7 [2@2Normal dCJmH :: Heading 1 $5CJ&& Heading 2.!. Heading 3CJ66 Heading 4 $56@b6 Heading 5 @&5CJ6@b6 Heading 6 @&>*CJ6@b6 Heading 7 @&6CJ6@b6 Heading 8 @&6CJ6 @b6 Heading 9 @&6CJ<A@<Default Paragraph Font6'@6Comment ReferenceCJ0@0 Comment TextCJ2@2TOC 8 X !2@2TOC 7 X !2@2TOC 6 X !2@2TOC 5@ X !22TOC 4p X !22TOC 3 X !22TOC 2 X !..TOC 1 X !&@&Index 7p&@&Index 6&@&Index 5& @&Index 48& @&Index 3& @&Index 2h" @"Index 1&(@& Line Number.!@. Index Heading! @1" Footer"2@22Header#d CJ<&@A<Footnote ReferenceCJEH8@R8 Footnote Text%dCJ2@b2 Normal Indent&6*@q6Endnote ReferenceH*,, ArtAuthors+,, ArtAbsText,CJ&1& ArtHeader->> NormalSans.dCJOJQJkH22 ArtFigTitle /$DQD ArtFigText0$$hd8xxCJ44ArtText1$dxCJ,", ArtRefText2$FFArtTextExample3hd<dCJ2B2 ArtPicture 4$dHRHNormalSansFixed5dCJOJQJkH@b@ table example6$dCJz >ZZ$e !+3:?HQZOAE$37A^]abdefF1CZ^^`cgh^_PXZf} "&46IQS m*x*z****---/////////000D0L0N0m00000000001333v9}99: : ::/:?:V:;.;0;;;;>>>>>>>>>?Z" 99" 99""   "" 99" :"""" 99 99:" 99::" 99R\_l99"99 seedplantbulb christeningclovebeanplantainflower2plant2hookanglingPI *D00:>Z [T *O00 :>ZKaren Hargrove!\\MSR\NLPUBLIC\PAPERS\VANDW95.DOCV  @HP LaserJet IIILPT1:HPPCL5MSHP LaserJet IIIHP LaserJet III@g,,@MSUDOHP LaserJet IIId HP LaserJet III@g,,@MSUDOHP LaserJet IIId  Q&Z@GTimes New Roman5Symbol3& Arial?&Lucida SansU5Lucida Sans Typewriter"V h  $%) GI #ILLV 0kTVLucy VanderwendeKaren HargroveOh+'0 $ @ L X dpxssLucy Vanderwendeducyucy Gudiss.dotwKaren Hargrovee2reMicrosoft Word 8.0@@]dxŹ@QV@QV) G՜.+,D՜.+,0 hp  MSFTitlIkT   Title 6> _PID_GUIDAN{49B5F4BB-C209-11D0-883A-3C8B00C10000}  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghiklmnopqrstuvwxyz{|}~Root Entry F ņ eQV Data jG1TableWordDocumentISummaryInformation(DocumentSummaryInformation8@HQV ߟQVVBA@HQV`QVThisDocument  _VBA_PROJECT; dirnJ  !"#$%&'()*+,-./0123456789:<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmopqrstuvwxz|}~xmo# (SLSS<N0{00020906-0000-0000-C000-000000000046}8(DO$HMExAttribute VB_Name = "ThisDocument" Bas0{00020P906-0C$0046} |CreatablFalse ^PredeclaIdTru "@ExposeTemplateD0eriv$eCustomiz2a_  *\G{000204EF-0000-0000-C000-000000000046}#3.0#9#C:\PROGRAM FILES\COMMON FILES\MICROSOFT SHARED\VBA\VBA332.DLL#Visual Basic For Applications*\G{00020905-0000-0000-C000-000000000046}#8.0#409#C:\Program Files\Microsoft Office\Office\MSWORD8.OLB#Microsoft Word 8.0 Object Library*\G{00020430-0000-0000-C000-000000000046}#2.0#0#C:\WINDOWS\SYSTEM\STDOLE2.TLB#OLE Automation*\G{E23FD16D-8A59-11D0-883A-3C8B00C10000}#2.0#0#C:\WINDOWS\SYSTEM\MSForms.TWD#Microsoft Forms 2.0 Object Library*\G{E23FD170-8A59-11D0-883A-3C8B00C10000}#2.0#0#C:\WINDOWS\TEMP\VBE\MSForms.EXD#Microsoft Forms 2.0 Object Library.E .`M *\G{2DF8D04C-5BFA-101B-BDE5-00AA0044DE52}#2.0#0#C:\PROGRAM FILES\MICROSOFT OFFICE\OFFICE\MSO97.DLL#Microsoft Office 8.0 Object Libraryh3 mThisDocument a3368cced*DoautoOpen b3368cced*Da8I :<I :<H+,P*WordkVBAWin16~Win32MacVANDW95Rstdole`MSFormsC ThisDocument< _EvaluateNormalOfficeuProject-Module1bautoOpen*MAIN, fileMacro1T globMacro MacroFileUv errCaughts` WordBasicFileSummaryInfoYUpdatedlg- DialogRecord CurValuesȤ DirectorymGFileNamejUCase5 Right$y1MacroFileName$J MacroName$ MacroCopyE FileSaveAsbyezRndWord\ wordNum.DocumentStatisticsLRndRValWords8StartOfDocument WordRightDocumentj` $D0 HProjectQ(@= l h3 " < rstdole>stdole( f\*\G{00020430-C 46}#2.0#0#C:\WINDOWS\SYSTEM\STDOLE2.TLB# AutomaPtionmMSForms>BSFErDmK/z pFE23FD16D-8A59-11D0-883A-3C8B00C1GF3.TWD#Microso(ft  Ob LibraDry$P0 r P7K*P@'P\V$BE(EX(.E .`M CxOfficDxO@u`ficBx,,2DF8D04C-5BFA-101B-BDE5@xAA@w42,PROGRAM FILES\MICROSOFT OFFICE\MSO97.DLL1# 81m-ThisDocu mentN2Ah@6sD@ecueenU@HB1P,BoT"B+BBB,Bo"B+B(aOpenJ2E@t@OpRnOMa!CThisDocumentThisDocumentID="{49B5F4BC-C209-11D0-883A-3C8B00C10000}" Document=ThisDocument/&H00000000 Name="Project" HelpContextID="0" CMG="6A68C14E1A521A521A521A52" DPB="CFCD641DACE710E810E810" GC="4240E9985D995D99A2" [Host Extender Info] &H00000001={3832D640-CF90-11CF-8E43-00A0C911005A};VBE;&H00000000 PROJECTwmy)PROJECT {4CompObjj  FMicrosoft Word Document MSWordDocWord.Document.89q