From vh@eel.cs.columbia.edu Fri Jun 9 13:30:00 2000 Date: Fri, 9 Jun 2000 15:29:54 -0400 (EDT) X-Authentication-Warning: eel.cs.columbia.edu: vh set sender to vh@eel.cs.columbia.edu using -f From: Vasileios Hatzivassiloglou To: wiebe@cs.nmsu.edu Subject: Adjectives Content-Length: 6800 Hi Jan, I finally got down to computing the gradability labels for all adjectives that I have human annotations for (453 of them). I am including this data below; it is sorted on adjective name, followed by +/- (gradable/non-gradable) according to the model/Cobuild, and +/- according to the automatic assignment. From vh@eel.cs.columbia.edu Wed Jun 14 13:01:56 2000 Date: Wed, 14 Jun 2000 15:01:53 -0400 (EDT) X-Authentication-Warning: eel.cs.columbia.edu: vh set sender to vh@eel.cs.columbia.edu using -f From: Vasileios Hatzivassiloglou To: wiebe@cs.nmsu.edu In-reply-to: <200006141640.KAA24259@dominica.cs.nmsu.edu> (message from Janyce Wiebe on Wed, 14 Jun 2000 10:40:08 -0600 (MDT)) Subject: Re: Adjectives Content-Length: 2705 Hi Jan, If the ML results are not better by using adjectives as predictors, then we cannot include them at this point. Perhaps they would get better if you run with the larger set of adjectives in the future. I agree with you on doing a journal paper that includes expanded versions of the COLING work and other experiments you have been doing on subjectivity. Regarding your specific questions: --- I have some questions about the adjectives: 1. may I put the lists of adjectives we used in the coling paper on my web page? I'll put them with the paper citation, so if anyone uses them, they will need to cite the paper. 2. I don't understand a couple things about the following excerpts from your message: I finally got down to computing the gradability labels for all adjectives that I have human annotations for (453 of them). I am including this data below; it is sorted on adjective name, followed by +/- (gradable/non-gradable) according to the model/Cobuild, and +/- according to the automatic assignment. Is the set of adjs in this email message a superset of the set we used in the coling submission? Or are these additional adjectives? COnsider, e.g., able + - This means that the manual annotation is + and the automatic annotation is -? The automatic labels were produced completely automatically, right, without requiring some prior manual annotation? ---- > 1. may I put the lists of adjectives we used in the coling paper on > my web page? I'll put them with the paper citation, so if > anyone uses them, they will need to cite the paper. Good idea. > Is the set of adjs in this email message a superset of the set we used in > the coling submission? Or are these additional adjectives? I have a total of 453 adjectives manually annotated for gradability. They are exactly the set of adjectives included in my last email. Earlier, I had trained on 353 of them and tested on 100, which I had sent you and we included in the COLING submission. Now, I calculated automatic labels for all 453. So the new data is a superset of the old, although the automatic labels assigned may be different (since a different training procedure was used). > COnsider, e.g., able + - > This means that the manual annotation is + and the automatic > annotation is -? Yes. > The automatic labels were produced completely automatically, > right, without requiring some prior manual annotation? Yes, in the sense that they are the result of an automatic procedure starting from some training (manually annotated) data. In this latest case, cross-validation was used, so eventually the manual label for all adjectives was used in some training sets and an automatic label was found for each adjective (in a run with no access to the manual label for that adjective). -------------------------------------------------------------------------- able + - academic - - acceptable + + accurate + + active + + actual - - additional - - adequate + - administrative - - advisory - - afraid + + after-tax - - aggressive + + agricultural - - alleged - - alternative - - ambitious + + angry + + annual - - antitrust - - appropriate + + assistant - - associate - - attractive + + australian - - automatic - - automotive - - aware + + back - - bad + + bearish + + big + + bitter + + black + - blue - + blue-chip - - brief + + bright + + broad + + bullish + + busy + + capable + + careful + + cautious + + certain + - cheap + + chemical - - chief - - civil - - civilian - - clear + + clinical - - close + + cold + + comfortable + + commercial - - comparable - - competitive + + complete - + complex + + composite - - comprehensive + + confident + + confidential + - congressional - - consecutive - - conservative + + considerable + - consistent + + constitutional - - controversial + + conventional + + convertible - - corporate - - correct - + cost-cutting - - costly + + covert + - creative + + criminal - - critical + + crucial + + cultural - - cumulative - - current - - daily - - dangerous + + dead - - deep + + definitive - - democratic + + dependent + + deputy - - different + + difficult + + diplomatic + - direct + + disappointing + + domestic - - dominant + + double - - dramatic + + due - - eager + + early + + easy + + economic - - educational + - efficient + + elaborate + + elderly + - electric - - electrical - - electronic - - emotional + + enormous - - environmental - - equal - + equivalent - - essential + - excellent - - excess - - excessive + + exchange-rate - - exclusive + + executive - - expensive + + experimental + - extensive + + extraordinary + - fair + + false - - familiar + + famous + + far + + fast + + favorable + + favorite - - federal - - few + + final - - financial - - fine + + firm + + first-quarter - - five-year - - fixed-rate - - flat + + flexible + + floating-rate - - formal + + former - - four-year - - fourth-quarter - - free + + frequent + + fresh + + friendly + + front - - full + + full-year - - fundamental + + further - - future - - generous + + giant - - global - - gold - - good + + grand + + great + + guilty + - happy + + hard + + healthy + + hefty + + high + + high-tech + - historic + - historical - - hot + + hourly - - huge + - human - - illegal - - immediate - + immune - - important + + impossible + - impressive + + improper + - inadequate + + independent + - individual - - industrial - - inevitable - - inflationary + - influential + + informal + - initial - - institutional - - intense + + interesting + + internal - - international - - interstate - - joint - - judicial - - key - - large + + late + + left - - legal - - legislative - - legitimate + + lengthy + + liberal + + light + + like - + likely + + little + + long + + long-distance - - long-term + - low + + lucrative + + main - - major + - mandatory - - many + + maximum - - medical - - metric - - middle - - military - - minimum - - minor + + moderate + + modest + + monetary - - monthly - - moral + - mortgage-backed - - municipal - - mutual - - narrow + + national - - nationwide - - near + + nearby - - necessary + - negative + + nervous + + net - - new + + nice + + nonprofit - - normal + - northern - - nuclear - - numerous + + obvious + + official - - offshore - - old + + one-time - - one-year - - only - - open - + operating - - optimistic + + ordinary + + original + - other - - outside - - outstanding + - over-the-counter - - overall - - overseas - - own - - past - - payable - - perfect - + permanent - - pharmaceutical - - plastic - - poor + + popular + + positive + + possible - - potential - - powerful + + practical + + pre-tax - - precious + + preferred - - preliminary + - present - - presidential - - previous - - primary - - prime - - principal - - prior - - private - - professional + + profitable + + prominent + + promising + + proper + - prospective - - protectionist + + public - - punitive - - quarterly - - quick + + quiet + + radical + + rapid + + rare + + raw - - ready - - real - - real-estate - - reasonable + + recent + + red - + regional - - regular + - regulatory - - relative + - reluctant + + remarkable + + residential - - responsible + + retail - - rich + + right - - risky + + rival - - routine + + rural + + safe + + scientific - - seasonal - - second-largest - - second-quarter - - secondary - - secret + - senior + - sensitive + + separate - - serious + + severe + + sexual - - sharp + + short + + short-term + + significant + + similar + - single - - six-month - - sizable + - skeptical + + slight + + slow + + sluggish + + small + + so-called - - social - - soft + + sole - - solid + + sophisticated + + sound + + southern - - special - - speculative + + square - - stable + + standard - - state-owned - - steady + + steep + + strategic - - striking + + strong + + subject - - subsequent - - subsidiary - - substantial + + successful + + sudden + - sufficient - - sure + + surprising + + tax-exempt - - taxable - - technological - - temporary - - thin + + third-quarter - - three-month - - three-year - - tight + + tiny + + top - - total - - tough + + traditional + + tremendous - - true - + two-year - - typical + + ultimate - - unable - - uncertain + + unchanged - - unclear + + unexpected + - unfair + + unique - - unlikely + + unprecedented - - unsecured - - unsolicited - - unsuccessful + - unusual + + upper - - urban - - useful + + usual - + valuable + + various - - vast - - vital + + volatile + + voluntary - - vulnerable + + weak + + wealthy + + weekly - - western - - white + - whole - - wholesale - - wide + + widespread + + willing + + world-wide - - worth - + wrong + - young + +