Linguistics 567, Spring 2005, Jonathan Pool Lab 3 (Esperanto) Preparation Vocabulary Nouns: king reĝo turnip rapo man (individual) homo man (male adult) viro staple_food pano water akvo dog hundo peace paco garden ĝardeno student lernanto tomato tomato Intransitive verbs: rave frenezi flow flui die morti walk marŝi rain pluvi Transitive verbs: ridicule moki find trovi enter eniri fear timi like ami Other verbs: go iri Verb that is normally used only intransitively: "morti" (die). Case Esperanto has overt case. Subjects have the nominative case. Direct objects of transitive verbs have the accusative case. Case applies to nouns, pronouns, adjectives, adverbs, and pronominal and adjectival "correlatives". Other contexts, including predicate phrases and prepositional objects, behave according to patterns not yet described here. No verb requires an exceptional case. Case is expressed morphologically with the ending -n. This ending is invariant. Case interacts with semantics. The accusative case can generally alternate with prepositions. But then one might not know which preposition has been replaced by an accusative-case inflection. Typically, the accusative alternation appears when the preposition would have designated a destination, a time, a duration, or a dimension. Case also interacts with case. It is generally considered marginally grammatical for a verb to have more than one accusative-case complement. For example, if a transitive verb is made causative, it has two implicit complements, the one that is made to act and the one that is acted on, and either if made overt can have the accusative case, but if both are overt then the speaker avoids making them both complements of the verb by reformulating the clause. Agreement Agreement types: person no number yes gender no case yes definiteness no Agreement constituents: subject-verb no object-verb no determiner-noun if uninflected determiner, no; if inflected determiner, yes adjective-noun yes Agreement paradigm: The adjective (or inflected determiner) and noun both take the -j ending if plural, or the -n ending if accusative, or the -jn endings if both plural and accusative. Test Suite for Case and Adjective-Noun Agreement frenezaj reĝoj lernantojn timas ;crazy-PL kings students-ACC fear-PRES ;'Crazy kings fear students.' reĝoj frenezaj lernantojn timas ;kings crazy-PL students-ACC fear-PRES ;'Crazy kings fear students.' reĝoj frenezajn lernantojn timas ;kings crazy-PL-ACC students-ACC fear-PRES ;'Kings fear crazy students.' reĝoj frenezaj frenezajn lernantojn timas ;kings crazy-PL crazy-PL-ACC students-ACC fear-PRES ;'Crazy kings fear crazy students.' reĝoj frenezajn frenezaj lernantojn timas ;kings crazy-PL-ACC crazy-PL-ACC students-ACC fear-PRES ;'Crazy kings fear crazy students.' ;(very rare word order) *reĝoj frenezanj lernantojn timas ;kings crazy-ACC-PL students-ACC fear-PRES *reĝoj frenezajn lernantonj timas ;kings crazy-PL-ACC student-ACC-PL fear-PRES reĝoj frenezajn timas lernantojn ;kings crazy-PL-ACC fear-PRES students-ACC ;'Kings fear crazy students.' ;(rare word order) *frenezj reĝoj lernantojn timas ;crazy-0-PL kings students-ACC fear-PRES *reĝj frenezaj lernantojn timas ;king-0-PL crazy-PL students-ACC fear-PRES Test Suite for Case and Determiner-Noun Agreement This test suite presupposes that inflected determiners are subject to the same syntactic rules as adjectives, there is at most 1 uninflected determiner per noun, and if present it must immediately precede the noun and all other premodifiers of the noun. timas tiajn reĝojn niajn lernantoj ĉiuj ;fear-PRES such-PL-ACC kings-ACC our-PL-ACC students all-PL ;'All students fear such kings of ours.' timas tiajn reĝojn niaj lernantoj ĉiuj ;fear-PRES such-PL-ACC kings-ACC our-PL students all-PL ;'All our students fear such kings.' timas la tiajn reĝojn niaj lernantoj ĉiuj ;fear-PRES the such-PL-ACC kings-ACC our-PL students all-PL ;'All our students fear the kings of that kind.' *timas tiajn la reĝojn niaj lernantoj ĉiuj ;fear-PRES such-PL-ACC the kings-ACC our-PL students all-PL lernantoj ĉies hundojn mokas ;students everybody's dogs-ACC ridicule-PRES ;'Students ridicule everybody's dogs.' *lernantoj hundojn ĉies mokas ;students dogs-ACC everybody's ridicule-PRES lernantoj ĉies flavajn hundojn kadukajn mokas ;students everybody's yellow-PL-ACC dogs-ACC decrepit-PL-ACC ;ridicule-PRES ;'Students ridicule everybody's yellow decrepit dogs.' *lernantoj flavajn ĉies hundojn kadukajn mokas ;students yellow-PL-ACC everybody's dogs-ACC decrepit-PL-ACC ;ridicule-PRES Test Suite for Argument Optionality la hundoj frenezas ;the dogs crazy-PRES ;'The dogs are going crazy.' frenezas ;crazy-PRES ;'It's crazy.' la reĝo eniras la ĝardenon ;the king enter-PRES the garden-ACC ;'The king is entering the garden.' la reĝo eniras ;the king enter-PRES ;'The king is entering.' *la reĝo la ĝardenon ;the king the garden-ACC Test Suite for Modification eĉ guto malgranda konstante frapante traboras la monton granitan ;even drop-N un-large constantly hitting-ADV through-bore-PRES ;the mountain-ACC granite-ADJ-ACC ;'Even a droplet, dripping continually, bores through ; the granite mountain.' *guto traboras malgranda monton ;drop bore-through-PRES small mountain-ACC *traboras monton guto granitan ;through-bore-PRES mountain-ACC drop granite-ADJ-ACC la frapanta guto traboras monton ;the hitting drop through-bore-PRES mountain-ACC ;'The dripping drop bores through a mountain.' *la frapante guto traboras monton ;the hitting-ADV drop through-bore-PRES mountain-ACC Test Suite for Polar Questions This test suite pretends that tag questions do not exist. ĉu la paco mortas ;whether the peace die-PRES ;Is peace dying?' ĉu mortas la paco ;whether die-PRES the peace ;Is peace dying?' ĉu la paco ne mortas ;whether the peace not die-PRES ;Isn't peace dying?' *la paco ĉu mortas ;the peace whether die-PRES *la ĉu paco mortas ;the whether peace die-PRES Test Suite for Imperatives mortu ;die-IMP ;'Die.' la hundo mortu ;the dog die-IMP ;'The dog shall die.' mortu la hundo ;die-IMP the dog ;'The dog shall die.' ĉu la hundo mortu ;whether the dog die-IMP ;'Shall the dog die?' pluvu ;rain-IMP ;'Let there be rain (It shall rain).' ĉu pluvu ;whether rain-IMP ;'Shall it rain.' Test Suite for Modals This test suite ignores verb derivation that accomplishes the same semantic effect as do modal verbs. It also finesses the question, whether any modal class(es) of verbs should be analyzed as existing and, if so, which. Finally, it assumes that the order of the 3 constituents (subject, modal verb, main verb) in intransitive modal sentences and the order of the 4 constituents (subject, modal verb, main verb, object) in transitive modal sentences are free (see Kalocsay & Waringhien, 1985, p. 365, for support). ;lernantoj = student-PL ;povas = can ;moki = ridicule-INF ;reĝojn = kings-ACC ;All sentences in this set = 'Students can ridicule kings.' lernantoj povas moki reĝojn lernantoj povas reĝojn moki lernantoj moki povas reĝojn lernantoj moki reĝojn povas lernantoj reĝojn povas moki lernantoj reĝojn moki povas povas lernantoj moki reĝojn povas lernantoj reĝojn moki povas moki lernantoj reĝojn povas moki reĝojn lernantoj povas reĝojn lernantoj moki povas reĝojn moki lernantoj moki lernantoj povas reĝojn moki lernantoj reĝojn povas moki povas lernantoj reĝojn moki povas reĝojn lernantoj moki reĝojn lernantoj povas moki reĝojn povas lernantoj reĝojn lernantoj povas moki reĝojn lernantoj moki povas reĝojn povas lernantoj moki reĝojn povas moki lernantoj reĝojn moki lernantoj povas reĝojn moki povas lernantoj lernantoj devas reĝojn moki ;student-PL must-PRES kings-ACC ridicule-INF ;'Students must ridicule kings.' reĝojn volas lernantoj moki ;kings-ACC want-PRES student-PL ridicule-INF ;'Students want to ridicule kings.' esperas studentoj reĝojn moki ;hope-PRES students kings-ACC ridicule-INF ;'Students hope to ridicule kings.' *lernantoj moki reĝojn ;student-PL ridicule-INF kings-ACC *lernantoj povas moki reĝoj ;student-PL can ridicule-INF kings Test Suite for Sentential Negation This test suite ignores nonsentential negation and doesn't include any grammatical examples of nonsentential negation as either positive or negative examples.) paco ne mortas ;peace not die-PRES ;'Peace isn't dying.' ne mortas paco ;not die-PRES peace ;'Peace isn't dying.' *paco mortas ne ;peace die-PRES not ne pluvas ;not rain-PRES ;'It isn't raining.' *pluvas ne ;rain-PRES not Test Suite for Coordination This test suite ignores the "kaj-kaj" ('both-and') construction. lernantoj amas panon kaj akvon ;students like-PRES bread-ACC and water-ACC ;'Students like bread and water.' lernantoj amas panon kaj tomatojn ĝardenajn ;students like-PRES bread-ACC and tomatoes-ACC garden-ADJ-PL-ACC ;'Students like bread and garden tomatoes.' ;'Students like garden bread and garden tomatoes.' ;(Ambiguous) lernanto trovis panon kaj tomaton ĝardenan ;student found bread-ACC and tomato-ACC garden-ADJ-ACC ;'A student found bread and a garden tomato. ;(Thus must be parsed so "garden" does not modify "bread") lernanto trovis panon kaj tomaton ĝardenajn ;student found bread-ACC and tomato-ACC garden-ADJ-PL-ACC ;'A student found garden bread and a garden tomato. ;(Thus must be parsed so "garden" modifies both "bread" and ;"tomato") *lernantoj amas pano kaj akvon ;students like-PRES bread and water-ACC *lernantoj amas panon kaj ;students like-PRES bread-ACC and *lernantoj amas kaj panon ;students like-PRES and bread-ACC la reĝo amas la marŝojn homan kaj hundan ;the king like-PRES the walks-N-PL-ACC human-ACC and canine-ACC ;'The king likes the human and canine walks. (One of each kind)' la reĝo amas la marŝojn homajn kaj hundan ;the king like-PRES the walks-N-PL-ACC human-PL-ACC and canine-ACC ;'The king likes the human and canine walks.' ;(Multiple human, single canine) la reĝo amas la marŝojn homan kaj hundajn ;the king like-PRES the walks-N-PL-ACC human-ACC and canine-PL-ACC ;'The king likes the human and canine walks.' ;(Single human, multiple canine) *la reĝo amas la marŝojn homajn kaj hunda ;the king like-PRES the walks-N-PL-ACC human-PL-ACC and canine reĝojn timas kaj mokas lernantoj ;kings-PL-ACC fear-PRES and ridicule-PRES students ;'Students fear and ridicule kings.' *reĝojn timas lernantoj kaj mokas ;kings-PL-ACC fear-PRES students and ridicule-PRES Package Selection Esperanto has case and agreement. The package chosen is case, adjective-noun agreement, and lexical rules. The test suite for this task consists of the "Test Suite for Case and Adjective-Noun Agreement" and the "Test Suite for Case and Determiner-Noun Agreement", set forth above in the "Preparation" section. Pronouns I decided to consider the personal non-reflexive and reflexive pronouns and the impersonal pronoun to be "pronouns" and to assume that none of them may take a determiner. Determiner Specification Common nouns can appear with or without a determiner, and pronouns can appear only without a determiner. In this respect, the revised grammar made correct predictions. However, determiners continued to be permitted to satisfy complement requirements, and this is incorrect. Case The enhancement of noun lexical entries with case constraints correctly eliminated all parses of intransitive sentences with accusative subjects and transitive sentences with accusative subjects or nominative direct objects. It also halved the number of parses of grammatical transitive sentences, by eliminating parses that had analyzed the subject as direct object and the direct object as subject. Incorrect parses remained whenever a determiner appeared in a transitive sentence, because, in addition to the correct parse, the determiner was also analyzed as a complement. In addition, I have discovered that nouns and pronouns and their phrases are incorrectly analyzed as complements, and there is no limit to the number of complements accepted in a verb phrase. Adjectives I defined the "adjective-lex" type with an unspecified value for POSTHEAD, believing that this would reflect the free adjective-noun word order in Esperanto. I was unsure of this, because I didn't clearly understand the statement about English, "We have both orders (head-adj and adj-head), but adjectives are always prehead". I have not noticed any parsing behavior that seems to contraindicate this decision. Agreement The enhancement of the lexical entries for inflected determiners made sentences with determiner-noun agreement parse and those with disagreement not parse. Transitive sentences with determiners continued also to be misparsed as if the determiners were verb or verb-phrase complements. Adjective-noun agreement and determiner-adjective-noun agreement appeared to be enforced correctly by the grammar. Lexical Rules I defined lexical rules to begin dealing with number and case inflection and agreement in Esperanto. I began with the observation that every uninflected noun, pronoun, determiner, adjective, or adverb stem can be analyzed as beginning its life at a particular level of a shared inflectional scale. The levels can be described as: 0. Ready for category inflection. 1. Ready for number inflection. 2. Ready for case inflection. 3. Ready for use. Some lexemes, such as the uninflectable definite article "la" ('the'), begin life ready for use. Others, such as the personal pronouns like "mi" ('I'), begin life ready for case inflection. Others, such as the inflectable determiner "tiu" ('that'), begin life ready for number inflection. And others, such as the noun stem "hund" ('dog'), begin life ready for category inflection. Wherever a stem begins on the scale, it proceeds only forward (toward readiness for use). Although the phonological/orthographic form taken by each inflection is identical for all lexeme and word types (part of speech), we still need separate lexical rules for particular types, for three reasons. One is that the types differ in the features that need to have values added as inflections take place. A second is that the semantics of a word of a final category can depend on its original category. The third is that progress along the scale is not always continuous: Any lexeme that is inflected for the adverb category then skips over number inflection and is next ready for case inflection. In order to assure that all required lexical rules are applied (even if with null affixation) before a lexeme is ready for use and no lexical rule is reapplied (see the warning about cyclicality of constant lexeme-to-lexeme rules in the Matrix file), I defined a NEEDINF feature and made it appropriate for "word-or-lexrule" so it could be a feature of both the input and the output. Its value indicates where on the inflectional scale the input and output are located (e.g., the input is ready for number inflection, and the output is ready for case inflection). I also defined the original lexemes as INFLECTED -, so they wouldn't be usable as words until this value is changed by a case-inflection rule, which is a lexeme-to-word rule. After debugging, I found the parsing fully in accord with expectations. The trees output by the LKB parser show each noun or inflected determiner in the appropriate place on a multinode nonbranching tree, whose nodes show the respective rules applied below and above the word. The defects identified earlier with respect to determiners analyzed as complements of verbs and node labels not culminating with "S", in the case of transitive sentences, remain to be addressed. I have implemented this approach so far on nouns, determiners, and adjectives. I haven't yet implemented it on verbs or adverbs. Head-Modifier Rules After I defined head-modifier rules, both transitive and intransitive sentences elicited correct parses and many incorrect ones. For example, the sentence "la junaj hundoj trovis tiun nigran panon" ('the young dogs found that black bread') has one correct parse, but generated an edge-limit error. The sentence "junaj hundoj trovis tiun nigran panon" ('young dogs found that black bread') also has one correct parse, but also generated an edge-limit error. The sentence "hundoj trovis tiun nigran panon" ('dogs found that black bread') also has one correct parse, but generated 16 parses. Adjectives were being accepted as complements of verbs. Also, the adjective-head rule was parsing adjective-noun constructions twice, once with the nouns analyzed as words and once with them analyzed as phrases. Constraining non-adjective head types to have empty MOD lists decreased the number of incorrect parses, making them few enough to eliminate the edge-limit errors with grammatical sentences. In the case of "hundoj trovis tiun nigran panon", the number decreased from 16 to 4. The only constraint that affected this difference was that on the determiner. When the determiner MOD constraint was absent, the determiner was being misanalyzed not only as a complement of the verb (as before), but also as the non-head daughter in a head-adj-int-phrase with the verb phrase "hundoj trovis" as the head daughter. Batch Testing Testing on this Lab's test suite yielded expected results. With particular exceptions, all grammatical sentences were correctly parsed, and additional incorrect parses occurred whenever it was possible to misanalyze a determiner or adjective as a verb or verb phrase's complement. The other prior error that recurred was the mislabeling of VP and S nodes in transitive sentences. Having further studied word order in Esperanto, I have adopted the prevailing position of the leading descriptive grammarians (who double as advisory stylists) that, in addition to the free order of the verb and its arguments, each of the primary arguments (subject and direct object) may also be freely split within the sentence, so that adjectives and the nouns they modify may be located in any order and at any distance, at least if there are no other complements of the verb except a direct object. The grammar fails to accept test sentences with separated constituents, such as "reĝoj frenezajn timas lernantojn" ('Kings fear crazy students'), with the order 'kings-crazy-fear-students'. Another new error is in the labeling of adjectives in the parse trees. They are labeled "ADV". The Lab 2 test suite continues to be parsed with the same success and the same previously reported errors as before. The Lab 1 test suite was in English, so doesn't seem applicable to this grammar. A technical error is that LKB apparently misreads the beginning of a test-suite file when it is encoded in UTF-8.