* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download writeup - Courses
Survey
Document related concepts
Serbo-Croatian grammar wikipedia , lookup
Zulu grammar wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
French grammar wikipedia , lookup
Portuguese grammar wikipedia , lookup
Chinese grammar wikipedia , lookup
Modern Hebrew grammar wikipedia , lookup
Esperanto grammar wikipedia , lookup
Preposition and postposition wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Yiddish grammar wikipedia , lookup
Turkish grammar wikipedia , lookup
Polish grammar wikipedia , lookup
Latin syntax wikipedia , lookup
Transcript
ANLP: Assignment 2 Andrew T. Fiore 29 September 2004 Grabbing sentences containing "help," "helps," "helping," and "helped" from the Treebank corpus, 100 documents. These searches find 34 distinct sentences with one or more of these words occurring. Here are my modified chunking rules. Following the rules are some sentences from the "help" set, chunked both by the basic rules and by my modified rules. (There are no line breaks in the rules, but they will probably wrap, depending on how you are viewing this file.) NP-Chunks: (<DT>?<RB>?)?((<NN.*|POS|CD|PRP\$|\$>|(<JJ.*|RBR>(<,>)?)))*(<NN.*|POS|CD|WP|PRP>)+ PP-Chunks: (<IN|TO>)+<NP>(((<,>)?<NP>)*(<,>)?<CC><NP>)? VP-Chunks: (<MD|TO>)?(<VB.*>)+(<IN>)?(<TO>)?(<VB.*>)*(<RP|RB.*>)?(<NP>|<PP>)*(<IN>)?(<RP|RB.*>)? Clearly, NP-chunks are the most complicated, followed by VP-chunks. The amount of variation in describing things in the world is tremendous, so NP-chunks can have complex and lengthy structure. I decided that appositives should each constitute their own NPs to keep chunks simple and relatively atomic. To improve accuracy, I added many tags that can sometimes act like adjectives (CD, PRP$, $, RBR) to the adjective portion the RegExp, and many tags that can act like nouns (CD, WP, PRP) to the noun portion. I also allowed for commas in NPs only after JJs and RBRs, to avoid overrunning the boundaries of the NP, as I saw in some test sentences with the basic rules. To the PP chunker I added the possibility of <TO> to the other prepositions <IN>, because in the Penn treebank tags, <TO> indicates not only the infinitive "to" but also the preposition. (See http://www.scs.leeds.ac.uk/amalgam/tagsets/upenn.html) Also, I added the possibility of compound objects of the preposition: one or more NPs connected by commas or, for the last NP in a series, a coordinating conjunction. The major modification to the VP chunking was to allow for prepositions (for verbs like "step out") and TO infinitive markers. See sample sentences below for some examples of how the prepositions are matched. Also, I added adverbs (RB.*), where before were only particles (RP), which serve largely the same function within VPs. COMPARISON SENTENCE #1 Here's a sentence chunked with the provided basic rules. The possessive pronoun "its" should be attached to the noun it modifies ("chairman"). The complicated verb phrase, "is stepping in to help turn...around," is only partially captured because of the presence of "in," a preposition that functions more as as part of the verb, "stepping in," and as such has no prepositional phrase of its own, and the <to/TO> infinitive indicator. Finally, the verb phrase at the end, "said," is not recognized because the rules don't allow for a VP with no direct object following the verb. Base: (S: (NP: <Magna/NNP> <International/NNP> <Inc./NNP>) <'s/POS> (NP: <chief/JJ> <financial/JJ> <officer/NN>) <,/,> (NP: <James/NNP> <McAlpine/NNP>) <,/,> <resigned/JJ> <and/CC> <its/PRP$> (NP: <chairman/NN>) <,/,> (NP: <Frank/NNP> <Stronach/NNP>) <,/,> <is/VBZ> <stepping/VBG> <in/IN> <to/TO> (VP: <help/VB> <turn/VB> (NP: <the/DT> <automotive-parts/JJ> <manufacturer/NN>)) <around/IN> <,/,> (NP: <the/DT> <company/NN>) <said/VBD> <./.>) With my new rules, the chunking is more correct: Andrew: (S: (NP: <Magna/NNP> <International/NNP> <Inc./NNP> <'s/POS> <chief/JJ> <financial/JJ> <officer/NN>) <,/,> (NP: <James/NNP> <McAlpine/NNP>) <,/,> <resigned/JJ> <and/CC> (NP: <its/PRP$> <chairman/NN>) <,/,> (NP: <Frank/NNP> <Stronach/NNP>) <,/,> (VP: <is/VBZ> <stepping/VBG> <in/IN> <to/TO> <help/VB> <turn/VB> (NP: <the/DT> <automotive-parts/JJ> <manufacturer/NN>) <around/IN>) <,/,> (NP: <the/DT> <company/NN>) (VP: <said/VBD>) <./.>) With my rules, the compound noun phrase with POS, "Magma International Inc.'s", is correctly associated with the NP ending in "officer." The possessive pronoun "its" is correctly associated with the noun it modifies, "chairman." "Resigned" still isn't identified as a VP, but that's because it is mis-tagged as JJ instead of VBD. The large VP is correctly identified as one chunk, including the preposition "around" at the end of the VP, which is part of the expression "turn around," so it doesn't belong in a separate PP. There are many examples of this in informal English, so I thought it was reasonable to add an optional IN to the end of a VP. This is predicated on the PP-chunker running BEFORE the VP-chunker, so that prepositions that are part of a legitimate PP will be tagged as such first. I made the direct object following a verb optional: (<NP>|<PP>)* This allows intransitive verbs and, as in this sentence, transitive verbs with a relocated object to be chunked correctly. COMPARISON SENTENCE #2 Base: BASE: (S: (PP: <As/IN> (NP: <a/DT> <Foster/NNP> <Corporate/NNP> <Parent/NNP>)) <,/,> <you/PRP> (VP: <will/MD> <experience/VB> (NP: <the/DT> <same/JJ> <joy/NN>)) (VP: <felt/VBD> (PP: <by/IN> (NP: <Robert/NNP> <Bass/NNP>))) <,/,> (NP: <Lewis/NNP> <Ranieri/NNP>) <,/,> (NP: <William/NNP> <Simon/NNP>) <and/CC> (NP: <others/NNS>) <,/,> <who/WP> (VP: <find/VBP> (NP: <ways/NNS>)) <to/TO> (VP: <help/VB> <troubled/VBN> (NP: <savings/NNS> <institutions/NNS>)) <and/CC> <their/PRP$> (NP: <employees/NNS>) <help/VB> <themselves/PRP> <./.>) In the chunking with the basic rules, the pronoun "you" isn't identified as a noun chunk. The PP beginning with "by" is linked with only the first of its four objects. The <who/WP> is not identified as a form of noun. The TO in "to help" is not included with the VP whose infinitive it introduces. Finally, the personal pronouns "their" and "themselves" are not treated properly as an adjective and a direct object, respectively. Andrew: (S: (PP: <As/IN> (NP: <a/DT> <Foster/NNP> <Corporate/NNP> <Parent/NNP>)) <,/,> (NP: <you/PRP>) (VP: <will/MD> <experience/VB> (NP: <the/DT> <same/JJ> <joy/NN>)) (VP: <felt/VBD> (PP: <by/IN> (NP: <Robert/NNP> <Bass/NNP>) <,/,> (NP: <Lewis/NNP> <Ranieri/NNP>) <,/,> (NP: <William/NNP> <Simon/NNP>) <and/CC> (NP: <others/NNS>))) <,/,> (NP: <who/WP>) (VP: <find/VBP> (NP: <ways/NNS>)) (VP: <to/TO> <help/VB> <troubled/VBN> (NP: <savings/NNS> <institutions/NNS>)) <and/CC> (NP: <their/PRP$> <employees/NNS>) (VP: <help/VB> (NP: <themselves/PRP>)) <./.>) My chunking correctly identifies "you" as a noun. It groups the compound objects of <by/IN> as one PP. It treats the pronoun <who/WP> as part of a NP. It correctly pairs <to/TO> with its VP. Finally, it properly treats "their" and "themselves" as adjective and direct object, respectively. COMPARISON SENTENCE #3 Base: (S: (NP: <The/DT> <show/NN>) <did/VBD> <n't/RB> (VP: <give/VB> (NP: <the/DT> <particulars/NNS>) (PP: <of/IN> (NP: <Mrs./NNP> <Yeargin/NNP>))) <'s/POS> (NP: <offense/NN>) <,/,> <saying/VBG> <only/RB> <that/IN> <she/PRP> (VP: <helped/VBD> (NP: <students/NNS>)) <do/VB> <better/RBR> (PP: <on/IN> (NP: <the/DT> <test/NN>)) <./.>) The basic chunking, by not allowing for the many possible positions and tag variants for adverbs, fails to chunk most of the VPs in this sentence. Also, the POS after "Mrs. Yeargin" is detached from her NP. Andrew: (S: (NP: <The/DT> <show/NN>) (VP: <did/VBD> <n't/RB>) (VP: <give/VB> (NP: <the/DT> <particulars/NNS>) (PP: <of/IN> (NP: <Mrs./NNP> <Yeargin/NNP> <'s/POS> <offense/NN>))) <,/,> (VP: <saying/VBG> <only/RB> (PP: <that/IN> (NP: <she/PRP>))) (VP: <helped/VBD> (NP: <students/NNS>)) (VP: <do/VB> <better/RBR> (PP: <on/IN> (NP: <the/DT> <test/NN>))) <./.>) With my rules, the VPs with adverbs are joined more appropriately: "didn't," "saying only that she," "helped students," "do better on the test." The final three VPs could plausibly be treated as one or two, but it depends on how fine-grained a chunking the user wants. Also, the possessive is correctly attached to "Mrs. Yeargin"; together, they act as an adjective modifying "offense." OCCURRENCES OF "HELP" Breakdown by syntactic structure: (see below for listing of VPs with "help") 1. help VB (infinite with implied TO, "help keep stock") (15 instances) 2. help NP (direct object, "help the kids") (10 instances) 3. helped PP (passive voice with prepositional phrase, "helped by strength") (2 instances) 4. help TO VB (infinitive, "helping to limit") (1 instance) 5. help (no object) (1 instance) 6. help PP (no object, but PP functions like infinitive, "help in saving money") (1 instance) The subjects of "help" are structurally very diverse. Sometimes the subject is a simple noun or pronoun, including personal pronouns, or perhaps the impersonal construction "it helps." On the other hand, since this corpus consisted of many news articles, I also saw elaborate noun phrases, usually representing organizations or abstract forces, as the subject. Semantic structure: (see below for listing of VPs with "help") "Help" occurs with semantically diverse words. Predictably, some occurrences, especially those with syntactic structure 2 from the above list (taking a direct object), involve providing aid of some sort. In several cases in my test sentences, the object was youth: kids, students, (indirectly) juvenile diabetes. Several of these occurrences come from the same article or series of articles, about a teacher who apparently gave illicit aid to some students on a test. With syntactic structures 1 and 4 (see above), the word "help" loses some of its meaning. In these cases, it carries the connotation not only of literally providing aid but also of incompleteness -- the subject is not performing the action thoroughly, or alone, or until completion, e.g.: "Policy makers regard the youth wage as helping to limit the loss of jobs from an increase in the minimum wage , but they have lately touted it as necessary to help impart job skills to entrants into the work force ." The subjects of "help" are diverse, as the word is common and somewhat generic. Both people and inanimate objects (often abstract events or conditions, e.g., "the subsequent flood of ... orders") occur as the subject. The objects of "help" are also diverse, but there are many occurrences of the young or underprivileged as objects of the word. Often the grammatical object of "help" is an infinitive phrase (with or without an explicit TO), so, semantically, its object is the notion of some action, event, or outcome. SEMANTIC LISTING will help meet increasing is stepping in to help turn the automotive-parts manufacturer around help build a larger version of its popular 767 twin-jet will help turn Southeast Asia into a more cohesive economic region was trying to help kids in an unfair testing situation wanted to help lift Greenville High School help the poor underprivileged child wanted to help could help save her teaching certificate could help treat juvenile diabetes help develop criminal cases help Georgia Gulf restructure help clear the myriad obstacles help in saving money will help keep a needy savings help troubled savings institutions help themselves was needed to help keep stock will help the overall market help it helping to limit the loss of jobs from an increase in the minimum wage help impart job skills helping her 'll be helping a neighborhood S&L in areas helped by strength in the defense capital goods sector helped students helped much by the announcement has helped underpin the dollar against the yen has helped lure investors have helped ratings at SYNTACTIC LISTING (VP: <will/MD> <help/VB> <meet/VB> <increasing/VBG>) (VP: <is/VBZ> <stepping/VBG> <in/IN> <to/TO> <help/VB> <turn/VB> (NP: <the/DT> <automotive-parts/JJ> <manufacturer/NN>) <around/IN>) (VP: <help/VB> <build/VB> (NP: <a/DT> <larger/JJR> <version/NN>) (PP: <of/IN> (NP: <its/PRP$> <popular/JJ> <767/CD> <twin-jet/NN>))) (VP: <will/MD> <help/VB> <turn/VB> (NP: <Southeast/NNP> <Asia/NNP>) (PP: <into/IN> (NP: <a/DT> <more/RBR> <cohesive/JJ> <economic/JJ> <region/NN>))) (VP: <was/VBD> <trying/VBG> <to/TO> <help/VB> (NP: <kids/NNS>) (PP: <in/IN> (NP: <an/DT> <unfair/JJ> <testing/NN> <situation/NN>))) (VP: <wanted/VBD> <to/TO> <help/VB> <lift/VB> (NP: <Greenville/NNP> <High/NNP> <School/NNP>)) (VP: <help/VB> (NP: <the/DT> <poor/JJ> <underprivileged/JJ> <child/NN>)) (VP: <wanted/VBD> <to/TO> <help/VB>) (VP: <could/MD> <help/VB> <save/VB> (NP: <her/PRP> <teaching/NN> <certificate/NN>)) (VP: <could/MD> <help/VB> <treat/VB> (NP: <juvenile/JJ> <diabetes/NN>)) (VP: <help/VB> <develop/VB> (NP: <criminal/JJ> <cases/NNS>)) (VP: <help/VB> (NP: <Georgia/NNP> <Gulf/NNP> <restructure/NN>)) (VP: <help/VB> <clear/VB> (NP: <the/DT> <myriad/JJ> <obstacles/NNS>)) (VP: <help/VB> <in/IN> <saving/VBG> (NP: <money/NN>)) (VP: <will/MD> <help/VB> <keep/VB> (NP: <a/DT> <needy/JJ> <savings/NNS>)) (VP: <help/VB> <troubled/VBN> (NP: <savings/NNS> <institutions/NNS>)) (VP: <help/VB> (NP: <themselves/PRP>)) (VP: <was/VBD> <needed/VBN> <to/TO> <help/VB> <keep/VB> (NP: <stock/NN>)) (VP: <will/MD> <help/VB> (NP: <the/DT> <overall/JJ> <market/NN>)) (VP: <help/VB> (NP: <it/PRP>)) (VP: <helping/VBG> <to/TO> <limit/VB> (NP: <the/DT> <loss/NN>) (PP: <of/IN> (NP: <jobs/NNS>)) (PP: <from/IN> (NP: <an/DT> <increase/NN>)) (PP: <in/IN> (NP: <the/DT> <minimum/JJ> <wage/NN>))) (VP: <help/VB> <impart/VB> (NP: <job/NN> <skills/NNS>)) (VP: <helping/VBG> (NP: <her/PRP>)) (VP: <'ll/MD> <be/VB> <helping/VBG> (NP: <a/DT> <neighborhood/NN> <S&L/NN>) (PP: <in/IN> (NP: <areas/NNS>))) (VP: <helped/VBN> (PP: <by/IN> (NP: <strength/NN>)) (PP: <in/IN> (NP: <the/DT> <defense/NN> <capital/NN> <goods/NNS> <sector/NN>))) (VP: <helped/VBD> (NP: <students/NNS>)) (VP: <helped/VBN> <much/RB> (PP: <by/IN> (NP: <the/DT> <announcement/NN>))) (VP: <has/VBZ> <helped/VBN> <underpin/VB> (NP: <the/DT> <dollar/NN>) (PP: <against/IN> (NP: <the/DT> <yen/NN>))) (VP: <has/VBZ> <helped/VBN> (NP: <lure/NN> <investors/NNS>)) (VP: <have/VBP> <helped/VBN> (NP: <ratings/NNS>) <at/IN>) SENTENCE LISTING The new plant , located in Chinchon about 60 miles from Seoul , will help meet increasing and diversifying demand forcontrol products in South Korea , the company said . Magna International Inc. 's chief financial officer , James McAlpine , resigned and its chairman , Frank Stronach , is stepping in to help turn the automotive-parts manufacturer around , the company said . Boeing Co. said it is discussing plans with three of its regular Japanese suppliers to possibly help build a larger version of its popular 767 twin-jet . Japanese money will help turn Southeast Asia into a more cohesive economic region . After numerous occurrences of questionable teacher help to students , Texas is revising its security practices . `` I was trying to help kids in an unfair testing situation , '' she says . '' Mrs. Yeargin says that she also wanted to help lift Greenville High School 's overall test scores , usually near the bottom of 14 district high schools in rankings carried annually by local newspapers . They found students in an advanced class a year earlier who said she gave them similar help , although because the case was n't tried in court , this evidence was never presented publicly . `` That pretty much defeats any inkling that she was out to help the poor underprivileged child , '' says Joe Watson , the prosecutor in the case , who is also president of Greenville High School 's alumni association . Mrs. Yeargin concedes that she went over the questions in the earlier class , adding : `` I wanted to help all '' students . She says she offered Mrs. Yeargin a quiet resignation and thought she could help save her teaching certificate . Medical researchers believe the transplantation of small amounts of fetal tissue into humans could help treat juvenile diabetes and such degenerative diseases as Alzheimer 's , Parkinson 's and Huntington 's . But they have obtained 8300 forms without court permission and used the information to help develop criminal cases . The offer follows an earlier proposal by NL and Mr. Simmons to help Georgia Gulf restructure or go private in a transaction that would pay shareholders $ 55 a share . But for small American companies , it also provides a growing source of capital and even marketing help . Partly to help clear the myriad obstacles facing any overseas company trying to penetrate Japan , tiny Candela turnedto Mitsui & Co. , one of Japan 's largest trading companies , for investment . The program not only offers a pre-approved car loan up to $ 18,000 , but throws in a special cash-flow statement to help in saving money . Your $ 15,000 will help keep a needy savings and loan solvent -- and out of the federal budget deficit . As a Foster Corporate Parent , you will experience the same joy felt by Robert Bass , Lewis Ranieri , William Simon and others , who find ways to help troubled savings institutions and their employees help themselves . Do n't wait -- a savings institution needs your help now ! But when the contract reopened , the subsequent flood of sell orders that quickly knocked the contract down to the 30-point limit indicated that the intermediate limit of 20 points was needed to help keep stock and stock-index futures prices synchronized . But Mr. Boesel of T. Rowe Price , who also expects 12 % growth in dividends next year , does n't think it will help the overall market all that much . The company , which recently said it lacked the profits and capital to pay dividends on its Series A convertible preferred stock , said it has hired an investment banker to help it raise additional cash . Policy makers regard the youth wage as helping to limit the loss of jobs from an increase in the minimum wage , but they have lately touted it as necessary to help impart job skills to entrants into the work force . Mrs. Yeargin 's extra work was also helping her earn points in the state 's incentive-bonus program . As a Foster Corporate Parent , you 'll be helping a neighborhood S&L in areas crucial to its survival . Policy makers regard the youth wage as helping to limit the loss of jobs from an increase in the minimum wage , but they have lately touted it as necessary to help impart job skills to entrants into the work force . Manufacturers ' backlogs of unfilled orders rose 0.5 % in September to $ 497.34 billion , helped by strength in the defense capital goods sector . The show did n't give the particulars of Mrs. Yeargin 's offense , saying only that she helped students do better on the test . The stocks of banking concerns based in Massachusetts were n't helped much by the announcement , traders said , because many of those concerns have financial problems tied to their real-estate loan portfolios , making them unattractive takeover targets . While market sentiment remains cautiously bearish on the dollar based on sluggish U.S. economic indicators , dealers note that Japanese demand has helped underpin the dollar against the yen and has kept the U.S. currency from plunging below key levels against the mark . Jay Goldinger , with Capital Insight Inc. , reasons that while the mark has posted significant gains against the yen as well -- the mark climbed to 77.70 yen from 77.56 yen late Tuesday in New York -- the strength of the U.S. bond market compared to its foreign counterparts has helped lure investors to dollar-denominated bonds , rather than mark bonds. The reruns have helped ratings at many of the 187 network affiliates and independent TV stations that air the shows .