This file contains notes on new or altered parsing policy originally written as guidelines for the annotators. ============================================================================== There are 3 new dash-tags being used in conjunction with Treebank II for bracketing Switchboard. (1) -UNF This tag marks unfinished constituents. It is added to the lowest constituent that can be labelled with confidence. ( (S (NP-SBJ it) (VP-UNF 's) , N_S)) ( (S (NP-SBJ-1 I) (VP happen (S (NP-SBJ *-1) (VP to (VP live (ADVP-LOC (ADVP not too far) away (PP-UNF from)))))) , N_S)) (2) -ETC This tag marks a special construction, which is extremely common in speech, in which a conjunct of the type `or anything', `and whatever', `and everything else' ends (usually) the sentence. Rather than use the UCP (unlike conjoined phrase) continually for this, we simply allow unlike conjoined phrases of this type, but tag the second conjunct -ETC. ( (S (NP-SBJ I) (VP have (NP no idea (SBAR (WHADVP-2 how long) (S (NP-SBJ-1 this) (VP is (VP supposed (S (NP-SBJ *-1) (VP to (VP (VP last (ADVP-TMP *T*-2)) or (NP-ETC anything)))))))))) . E_S)) (3) In Treebank II, imperatives are not specially marked, except by the presence of an empty subject (NP-SBJ *). In Switchboard, imperatives will be marked with the dash-tag -IMP on the S label, and main clauses with empty subjects will be reserved for main clauses with no (i.e. small pro) subjects. Imperative: ( (S-IMP (NP-SBJ *) (VP let (S (NP-SBJ 's) (VP do (NP (NP something) (ADVP else))))) . E_S)) Subjectless clause: ( (S (NP-SBJ *-1) (ADVP probably) (VP need (S (NP-SBJ-1 *-2) (VP to (VP try (S (NP-SBJ-2 *) (VP to (VP get (ADVP-CLR back) (PP-PRD on (NP the topic))))))))) E_S)) =============================================================================== VP or S? In general, we've been putting in an (NP-SBJ *) at top level whenever there is at least a VP actually present. I think we should distinguish cases that are missing just a subject and cases which are also missing one or more auxiliaries. So if ONLY the subject is missing then put in (NP-SBJ *). But if an auxiliary is missing as well, then make it VP at the top level. In cases where the verb is simple, i.e. there are no auxiliaries, and the subject is missing, the top level is S. ( (S (NP-SBJ *) '(it/that) might make you feel better' (VP Might (VP make (S (NP-SBJ you) (VP feel (ADJP-PRD better))))) E_S)) ( (S (INTJ Um) '(that) sounds good' . (NP-SBJ *) (VP Sounds (ADJP-PRD good)) . E_S)) ( (VP Think '(do you) think so?' (ADVP so) ? E_S))) ( (VP Supposed '(she was) supposed to be ...' (S (NP-SBJ *-1) (VP to (VP be (NP-PRD (NP good , (INTJ uh) , recommended person) (PP from (NP the church)))))) N_S)) ( (VP be '(you are going) to be one of the ...' (NP-PRD (NP one) (PP of (NP (NP the last ones) (SBAR (WHNP 0) (S (NP-SBJ *) (VP-UNF to (VP let (VP go)))))))) (ADVP anyway) . E_S)) This should apply to all predicates really, so ( (ADJP-PRD Worried '(are you) worried that ...' (SBAR that (S (NP-SBJ-1 they) (VP 're not (VP going (S (NP-SBJ *-1) (VP to (VP get (NP enough attention)))))))) ? E_S)) =============================================================================== SOME PROBLEMATIC DYSFLUENCY STUFF The use of UNF There is one and only one -UNF per sentence. This -UNF goes on the lowest unfinished constituent. If the lowest constituent happens to be complete, the UNF will go on a higher constituent, but it should still be on the lowest unfinished constituent. S-UNF covers the following cases: A conjunction alone (filler type INTJs allowed but not assessors) ( (S-UNF so N_S)) ( (S-UNF and (INTJ uh) N_S)) ( (FRAG and (INTJ yeah) E_S)) Any constituent up to but not including the VP (as long as all are complete) ( (S-UNF and (NP-SBJ John) (ADVP really) N_S)) What to put at top level. Top level labelling apart from S, SBARQ, SQ, SINV, is only used when the constituent is complete (or possibly if somehow it is absolutely clear from context that all they intended to say was that constituent). Where do you live? ( (NP Santa Barbara E_S)) In all other cases assume that the incomplete constituent is the start of a new S. If the constituent is an NP, then assume (again unless you have really good reason not to) that it is the subject. If it is possible to assume that it is complete (e.g. 'that', 'this' which could be complete or not) do so. An extreme example is a bare 'the' at the beginning of a token ( (S (NP-SBJ-UNF the) N_S)) I think, because SBARs not uncommonly stand alone in this data, that we can have top level unfinished SBARs as well. ( (SBAR-PRP-UNF because N_S)) ( (SBAR-ADV-UNF if N_S)) But if the SBAR is preceded by a conjunction, then it must be S. ( (S and (SBAR-ADV-UNF if) N_S)) Separating stuff Every assesor (yeah, uh-huh, right, exactly, etc) gets its own token, so separate any that come to you joined. Don't separate off 'because' clauses, just follow what the dfl person did (unless they clearly go with the following clause rather than the one they're in). =============================================================================== ATTACHMENT LEVELS The general rule is 'when in doubt attach high'. See below for predicates with 'be'. Don't assume anything in this part applies to 'be'. When trying to choose between making a PP a modifier of an NP or putting it at VP level: Since PPs inside NPs are essentially reduced relatives, try putting 'which is/was/etc' between them. If you get a *RESTRICTIVE* relative clause which makes sense, then the PP goes inside the NP. If not, it goes at VP level. If it is doubtful, put it at VP level. Although putting in 'which is' in the first one below isn't bad, the result is not a restrictive relative clause. In the second case, it is restrictive and should be done as a reduced relative. Never adjoin two PPs like this together. ( (S (NP-SBJ Ann) (VP lives (PP-LOC in (NP Ardmore)) (PP-LOC near (NP Philadelphia))))) ( (S (NP-SBJ Ann) (VP lives (PP-LOC in (NP (NP the Ardmore) (PP-LOC near (NP Philadelphia))))))) (not the one near someplace else) Sometimes the level makes a big difference. In the first one below, it's the 'presence' which is in the Gulf, and the location of the protest is unspecified. In the second, the protesting took place in the Gulf. ( (S (NP-SBJ they) (VP protested (NP (NP the presence) (PP-LOC in (NP the Gulf)))))) ( (S (NP-SBJ they) (VP protested (NP the presence) (PP-LOC in (NP the Gulf))))) Places where you need to use adjoined PPs are not too common. All the following type cases should be done as separate PPs, not adjoined (all these are taken from earlier files where the PPs were adjoined). ( (S (NP-SBJ they) (VP live (PP-LOC up (ADVP there)) (PP-LOC in (NP the mountains))))) # /nldb/parstexts/swbd/00/sw_0072_3876.prd (TOP (S (INTJ Now) , (NP-SBJ I) (VP work (PP-LOC at (NP J C PENNY)) (PP-LOC at (NP their corporate headquarters))) . E_S)) # /nldb/parstexts/swbd/01/sw_0123_3186.prd (TOP (FRAG (INTJ Well) , (NP (NP thanks) (PP for (S-NOM (NP-SBJ *) (VP calling))) (PP for (S-NOM (NP-SBJ *) (VP helping (NP us) (PRT out))))) . E_S)) # /nldb/parstexts/swbd/02/sw_0213_2285.prd (TOP (S (PRN (S (NP-SBJ I) (VP mean))) (SBAR-ADV if (S (NP-SBJ the parents) (VP are n't (VP supplying (NP it))))) , (NP-SBJ-1 they) (VP 've (VP got (S (NP-SBJ *-1) (VP to (VP get (NP it) (PP from (NP (NP someone) (ADVP else))) (PP from (NP the schools))))))) , E_S)) # /nldb/parstexts/swbd/02/sw_0222_2676.prd (TOP (S (NP-SBJ *) (VP Lived (PP-LOC up (ADVP north (PP of <---<<< or adjoined? (NP Los Angeles)))) (PP-LOC in (NP (NP Thousand Oaks area) , (SBAR (WHADVP-1 where) (S (NP-SBJ the Cowboys) (VP have (NP their training camp) (ADVP-LOC *T*-1))))))) ,)) # /nldb/parstexts/swbd/02/sw_0222_2676.prd (TOP (S (NP-SBJ they) (VP have (NP some (ADJP pretty nice) weather) (PP-LOC out (ADVP there)) (PP-LOC in (NP Los Angeles))) .)) # /nldb/parstexts/swbd/02/sw_0226_3081.prd (TOP (FRAG (SBAR-ADV if (S (NP-SBJ you) (VP leave (NP the pit) (PP-LOC in (NP the bowl)) (PP with (NP the thing))))))) Cases where you need to adjoin PPs Range (but not endpoints) Note that masses of endpoint cases are being done as adjoined. Make sure you understand the difference here. (FROM MANUAL) Where a range is indicated, {\it from...to...}\/ is annotated as a complex (conjoined) PP; where two end points are indicated, {\it from}\/ and {\it to}\/ are annotated as separate (nonconjoined) PPs. The distinction is made using the following test: if the order of the PPs in question can be reversed, then they constitute endpoints, and if not, they constitute a range. Note that {\it from...to...}\/ ranges in determiner position are called QP, as in the example {\it from 10 to 15 monkeys}\/ above, on page~\pageref{list:NPmod:measure/amount:QP:monkey-ex}. The following examples contain nouns modified by ranges/endpoints. range: (NP (NP a number) (PP (PP from (NP 2)) (PP to (NP 32)))) (NP (NP excursions) (PP (PP from (NP studio)) (PP to (NP studio)))) (VP varied (PP (PP from (NP 30)) (PP to (NP 53 mg.)))) endpoints: (NP (NP the transition) (PP from (NP vinyl records)) (PP to (NP compact discs))) (VP went (PP-DIR from (NP Paris)) (PP-DIR to (NP Dakar))) (VP went (PP-DIR from (NP general)) (PP-DIR to (NP specific terms))) With 'through' (NP (NP a number) (PP (PP from (NP 0)) (PP through (NP 6)))) As a general rule, when {\it through}\/ is not in construction with another preposition and occurs between two like categories, such as NPs, it is annotated as a conjunction: (NP numbers 4 through 9) (NP (NP the file mode number) (PRN *LRB* (NP 0 through 6) *RRB*)) When there is a choice of VPs to attach adjuncts to, first go by the meaning. In the following two similar sentences, the level of the SBAR differs based on what the SBAR is giving the reason for: John's coming in the first one, or my believing in the second one. Note that in the second one you can omit (or better replace with 'it') the SBAR 'John is coming' and the sentence still makes sense, whereas in the first one it doesn't. Always try to attach high first, omitting or pronominalizing the in-between bits of the clause to see if it works. If it doesn't then move it downward until it does. When in doubt, attach high. ( (S (NP-SBJ I) (VP believe (SBAR 0 (S (NP-SBJ John) (VP is (VP coming (SBAR-PRP because (S (NP-SBJ Mary) (VP told (NP him) (VP to (VP *?*)))))))))))) ( (S (NP-SBJ I) (VP believe (SBAR 0 (S (NP-SBJ John) (VP is (VP coming)))) (SBAR-PRP because (S (NP-SBJ Mary) (VP told (NP me) (ADVP so))))))) The case with 'be' is somewhat different, because anything can be a predicate. When there is more than one potential predicate, they should be adjoined when either can equally well act as predicate by itself and they are of the same category and type (i.e. both PPs and both LOC; this is the most common type, I think). # /nldb/parstexts/swbd/01/sw_0151_2772.prd (TOP (S (NP-SBJ we) (VP 're (PP-LOC-PRD (PP over (PP in (NP the western edge))) (PP in (NP the mountains)))) . E_S)) # /nldb/parstexts/swbd/02/sw_0220_2549.prd (TOP (S So , (NP-SBJ you) (VP were (PP-LOC-PRD (PP out (ADVP there)) (PP in (NP San Francisco)))) ? E_S)) If the constituents differ in category or type, do not adjoin them. ( (S (NP-SBJ I) (VP was (PP-LOC-PRD in (NP the country)) (PP-TMP in (NP the winter))))) If there is doubt just assume the first one is the predicate and treat the other as an adjunct. =============================================================================== ADJECTIVAL VERSUS VERBAL PASSIVES The distinction between adjectives and past participles is often very difficult to make. There are a number of tests that you can use to decide. Be sure to apply these tests to the entire sentence containing the word that you are unsure of, not just the word in isolation, since the context is important in determining the part of speech of a word.\ (From the TAGGING MANUAL) A word is an adjective: if it is gradable---that is, if it can be preceded by a degree adverb like 'very', or if it allows the formation of a comparative. EXAMPLE: He was very surprised/JJ. He was more surprised/JJ than she was. if there is a corresponding {\ems un-} form with the opposite meaning. EXAMPLE: a hurried/JJ meeting; an unhurried/JJ meeting Be sure to check whether there is a corresponding verb beginning with 'un-'. If there is, you cannot rely on this test to determine whether the word in question is an adjective or a participle, and you will have to use the other tests. EXAMPLE: Your shoelace has been untied/JJ ever since we started. I know---it got untied/VBN by accident. When applying the 'un-' test, be sure to take the entire context into account. For instance, 'armed' can be either a JJ or a VBN, depending on its context. EXAMPLE: We need an armed/JJ guard. (cf. We need an unarmed guard.) Armed/VBN with only a knife, ... (cf. *Unarmed with only a knife, ... ) if the word occurs in construction with 'be', and 'be' could be replaced by 'become', 'feel', 'look', 'remain', 'seem' or 'sound'. EXAMPLES: He became interested/JJ. He felt interested/JJ. He looked surprised/JJ. He remained surprised/JJ. He seemed surprised/JJ. He sounded surprised/JJ. However, if the complement of any of the verbs listed above is modified by a 'by'-phrase, it should be tagged as a participle (VBN) rather than as an adjective (JJ). EXAMPLE: He remains guided/VBN by these principles. if the word occurs in construction with ' keep'. EXAMPLES: They should be kept well watered/JJ. if it refers to a (resultant) state rather than to a (specific) event. EXAMPLES: At the time, I was married/JJ. I was mistaken/JJ (= wrong) the other day. a mistaken/JJ decision if a collocation of the form ``X-ed N'' cannot be paraphrased as ``N that has been X-ed.'' EXAMPLES: a decided/JJ advantage; *an advantage that has been decided a grown/JJ woman; *a woman that has been grown married/JJ life; *life that has been married worried/JJ faces; *faces that have been worried A word is a past participle (VBN): if it can be followed by a 'by' phrase. If this criterion clashes with the possibility of inserting a degree adverb, tag the word as an adjective (JJ), not as a participle (VBN). EXAMPLES: He was invited/VBN by some friends of hers. He was very surprised/JJ by her remarks. if it refers to an (specific) event rather than to a (resultant) state. EXAMPLES: I was married/VBN on a Sunday. \\ I was mistaken/VBN for you the other day. \\ a case of mistaken/VBN identity if the word occurs in construction with 'be', and 'be' could be replaced by 'get', but not by 'become'. EXAMPLE: I was married/VBN on a Sunday. (cf.\ I got married, *I became married) =============================================================================== SEZ POLICY Use SEZ to mark direct speech wherever it occurs. Since there are few quotation marks you'll have to use your judgement in some cases about what is direct or indirect. Some tipoffs for direct speech are changes in the pronouns or tense of the verb, subject-verb inversion in questions (indirect questions are not inverted), a comma following the verb (try saying it to yourself with a heavy pause between the verb and the complement; if you can't (in context), it is not direct speech), the use of interjections, especially sentence introducing ones like 'well' 'boy' 'God' 'yeah', 'now', 'see', but also just interjections in general. ( (S And (NP-SBJ I) (ADVP just) (VP said (S-SEZ (NP-SBJ this) (VP is <<--------<<< note tense change (ADJP-PRD terrible)))) . E_S)) Note that all non-traditional verbs of expression (be like, go, etc) always introduce direct speech. ( (S (NP-SBJ I) (VP was (PRT like) , (SBARQ-SEZ (INTJ God) , (WHNP-1 how much) (SQ were (NP-SBJ those bottles) (NP-PRD *T*-1) (PRN (S (NP-SBJ you) (VP know)))))) ? E_S)) We are going to count signs as direct speech. (NP-PRD (NP a (INTJ uh) , (INTJ uh) , (ADJP hastily erected) sign) (VP saying (NP-SEZ baby milk factory))) Direct speech can occur almost anywhere. Wherever it occurs, label it SEZ ( (S (NP-SBJ they) (VP get (NP-ADV very much) , (S-SEZ (INTJ well) (PRN , (S (NP-SBJ you) (VP know)) ,) (NP-SBJ it) (VP 's (ADVP just) (NP-PRD a big faceless corporation)))))) PLACES NOT TO USE SEZ The general strategy for bracketing non-sentential stuff after a verb of expression that is not direct speech is to use FRAG. Compare: ( (S (NP-SBJ he) <<----<<< no indication of direct speech: default indirect (VP said (FRAG (NP hot dogs))))) ( (S (NP-SBJ he) (VP said (FRAG-SEZ (INTJ oh) <<----<<< indications of direct speech (INTJ boy) (NP hot dogs))) !)) ( (S (NP-SBJ he) (VP said , <<----<<< slightly weak indications of direct speech (but possible (NP-SEZ hot dogs)) with context) !)) (1) after expressions like 'I'm wanting to say' 'I would say' 'I'll say' 'I wouldn't say' use FRAG not SEZ. ( (S (NP-SBJ-1 I) (VP 'm (VP wanting (S (NP-SBJ *-1) (VP to (VP say (FRAG (NP Raleigh Durham))))))) E_S)) (2) with 'mean' ( (S (FRAG-TPC-1 (PP From (NP (NP the lack) (PP of (NP stimulation))))) , (NP-SBJ you) (VP mean (FRAG *T*-1)) . E_S)) (3) traces after 'say' are not dash-tagged. these are like the cases above, 'she said that, he said something' ( (S (PRN (S (NP-SBJ I) (VP mean)) ,) (NP-SBJ it) (VP sounds (ADJP-PRD terrible (SBAR (WHNP-1 0) (S (NP-SBJ *) (VP to (VP say (NP *T*-1))))))) , E_S)) =============================================================================== *ATTACHMENT SITE OF RCS AND OTHER THINGS* To decide where to attach RCs (and other things) in NPs with adjoined PPs try leaving out the PP sentence still sounds good/makes sense: high sentence sounds rotten/doesn't make sense: low *CO-INDEXING* Empty subject is non-referential: no indexing Empty subject is co-referential with a non-adjunct NP (one not inside a PP): co-index EXCEPT when the clause containing the empty subject is inside an NP *SBAR 0* Complement is direct speech: do not use (SBAR 0). tipoffs for direct speech: changes in the pronouns changes in tense of the verb a comma following the verb try saying it to yourself with a heavy pause between the verb and the complement; if you can't (in context), it is not direct speech the use of interjections, especially sentence introducing ones like 'well' 'boy' 'God' 'yeah', 'now', 'see', but also just interjections in general default is indirect speech When speech is not involved, put in the (SBAR 0) if you can felicitously insert 'that'. Current exception (or did we get rid of this one too?): '(S (S-1 John did it, ) he said (SBAR 0 (S *T*-1))) *POSITION OF TRACES* Complement traces: immediately following verb (including with PRT) exception: double object verbs: indirect object trace first direct object trace second shared traces go in each complement (yes?) Adverbial traces: as far to the end of the clause as possible default level is the *matrix* verb shared traces at the same level go at conjunction level *DASH-TAGS* CLR: abolished LOC: restrict to physical location test: try paraphrasing with 'located in' trace of 'where' is LOC only for physical location TMP: for points of time, frequency of events, or stretches of time test for stretches of time: try paraphrasing with 'during' NEED A LIST OF ADVERBS HERE MNR: trace of 'how NEED MORE HERE ADV: all NPs, Ss and SBARs that are not complements must have a dash-tag if no more specific tag is appropriate, use ADV NOM: gerunds (-ing phrases) when acting as subject or object of preposition free relatives always infinitives never SEZ: direct speech in any context (for more details, see SEZ page) *DYSFLUENCY* see dysfluency page *MISCELLANEOUS* 'much' is primarily an adjective (NP (ADJP so much) controversy) All pre-verbal adverbs go outside the VP, even manner. 'a year and a half' (like 'a year or so') (NP a year (QP and a half)) 'not' can be a conjunction and goes at conjunction level in this case not inside the second NP ( (S (NP-SBJ he) (VP bought (NP (NP hot chocolate) not (NP bread))))) 'so' should be bracketed as it comes to you; if bare, as a conjunction, if in INTJ brackets as INTJ. Don't try to figure out the distinction, just follow what the dfl people did. 'plus' at the beginning of a sentence is a conjunction and is therefore unbracketed. 'even if' is a two-word complementizer. *Adjectival vs. verbal passives Stative reading: ADJP Eventive reading: VP *Position of INTJs and PRNs, punctuation and (RS ]) Finished sentence: PRN or INTJ goes at the matrix verb level (but see below) Unfinished sentence: PRN or INTJ goes inside the constituent marked UNF Always put PRNs and INTJs at the highest possible level (between SBAR and S this means at the SBAR level) Final punctuation: final punctuation goes at top level (this overrides the rule that commas go inside PRN brackets) Treat (RS ]) as punctuation. *Final INTJs, conjunctions and 'I mean' Split off from the end of a sentence: 'well', 'I mean', 'so', 'uh' Join 'well', 'I mean', 'so' to next sentence or make S-UNF if nothing follows Join 'uh' to next sentence or make top-level INTJ if nothing follows Do not split off 'you know' *S vs. VP Only subject missing: snipastar Subject and auxiliary missing: VP Only auxiliary missing: FRAG (you see that cute dog) <-- swbd only not nursing *Previously S-CLR Infinitival S-CLR --> S-PRP Resultative adjective S-CLR --> S (i.e. small clause) *DIR Only with verbs Restricted to physical movement from one point to another *DIR/LOC with 'be' be + to X = DIR-PRD be + there = LOC-PRD I'm from X = PRD *Common Temporal Adverbs* Always temporal: sometimes, always, now, right now, never, before, often, later, nowadays, afterwards, forever, lately, right away, already, ever, anymore, usually. Never temporal: again Sometimes temporal: yet, still, then For others, use your judgement. Temporal marks point in time, duration of time or frequency. *Possible Particles* The following words can be labelled PRT (when they appear in the appropriate context). Everything else that you might want to label PRT should be labelled ADVP. about, across, down, in, off, on, out, over, through, up, back, away, like (with 'be' introducing speech) *Miscellaneous go (ADVP-DIR home) stay (ADVP-LOC home) MNR-PRD is a legal label for a trace 'somewhere', 'everywhere', etc are ADVP don't forget QP on 'a thousand', 'a hundred', etc. =============================================================================== *Position of INTJs and PRNs at the end of finished and unfinished sentences If a sentence is FINISHED the PRN or INTJ goes at the matrix verb level. # /nldb/parstexts/swbd/00/sw_0002_4330.prd (TOP (S and (EDITED (RM [) (NP-SBJ-UNF tha-) , (IP +)) (NP-SBJ they) (RS ]) (VP did n't (VP find (NP (NP any problem) (PP with (NP that))) (PRN , (S (NP-SBJ you) (VP know))))) , E_S)) # /nldb/parstexts/swbd/00/sw_0002_4330.prd (TOP (S and (NP-SBJ I) (VP have (NP-TMP many a time) (VP called (NP-1 him) (S-PRP (NP-SBJ-2 *-1) (VP to (VP come (S (NP-SBJ *-2) (VP get (NP me)))))) (PRN , (S (NP-SBJ you) (VP know))))) . E_S)) # /nldb/parstexts/swbd/00/sw_0002_4330.prd (TOP (S (EDITED (RM [) And , (IP +)) (INTJ uh) , (INTJ uh) , but , (RS ]) (EDITED (RM [) (NP-UNF y-) , (IP +)) (PRN (S (NP-SBJ you) (VP know)) ,) (RS ]) (NP-SBJ-1 they) (VP do n't (VP think (ADVP-TMP twice) (PP-CLR about (S-NOM (NP-SBJ *-1) (VP serving (NP beer) (PP-MNR by (NP the keg))))) . (PRN (S (NP-SBJ You) (VP know))))) , E_S)) # /nldb/parstexts/swbd/00/sw_0001_4325.prd (TOP (S (EDITED (RM [) (S (NP-SBJ I) (VP-UNF guess)) (IP +)) (NP-SBJ I) (VP guess (RS ]) (SBAR 0 (S (NP-SBJ we) (VP can (VP start)))) . (INTJ Uh)) , E_S)) # /nldb/parstexts/swbd/00/sw_0001_4325.prd (TOP (S (NP-SBJ that) (VP 's (NP-PRD (NP something) (SBAR (WHNP-1 0) (S (NP-SBJ I) (VP 've (VP considered (NP *T*-1)))))) . (INTJ Uh)) , E_S)) # /nldb/parstexts/swbd/00/sw_0002_4330.prd (TOP (S (NP-SBJ-1 we) (VP 're not (VP being (VP tested (NP *-1) (PP for (NP drugs)) (ADVP at all) , (INTJ uh))))) , E_S)) If the sentence is UNFINISHED, then put the INTJ or PRN inside the constituent marked UNF # /nldb/parstexts/swbd/00/sw_0002_4330.prd (TOP (S (PRN (S (NP-SBJ I) (VP mean)) ,) (NP-SBJ I) (VP think , (PRN (S (NP-SBJ you) (VP know)))) , N_S)) # /nldb/parstexts/swbd/00/sw_0001_4325.prd (TOP (S so (NP-SBJ there) (VP are (NP (NP a lot) (PP of (NP (NP people) (SBAR (WHNP-1 who) (S (NP-SBJ *T*-1) (VP babysit (PP-LOC in (NP their homes))))) (SBAR-UNF (WHNP that) , (PRN (S (NP-SBJ you) (VP know)))))))) , N_S)) # /nldb/parstexts/swbd/00/sw_0001_4325.prd (TOP (S but , (INTJ uh) , (NP-SBJ *) (VP seems (PP-UNF like , (PRN (S (NP-SBJ you) (VP know))))) , N_S)) # /nldb/parstexts/swbd/00/sw_0001_4325.prd (TOP (S (NP-SBJ You) (VP might (VP-UNF try , (INTJ uh))) , E_S)) # /nldb/parstexts/swbd/00/sw_0001_4325.prd (TOP (S-IMP and (NP-SBJ *) (VP see (SBAR if (S-UNF (NP-SBJ it) , (INTJ uh)))) , N_S)) # /nldb/parstexts/swbd/00/sw_0001_4325.prd (TOP (SQ Is (NP-SBJ it) (PP-PRD-UNF like , (INTJ uh)) , N_S)) Where do PRNs go when between SBAR and S? ( (S And (EDITED (RM [) (S (NP-SBJ I) (VP-UNF am)) , (IP +)) (NP-SBJ-1 I) (VP am (RS ]) (ADJP-PRD afraid (S (NP-SBJ *-1) (VP to (VP go (PP-DIR down (ADVP there)))))) (SBAR-PRP because (S (PRN , (S (NP-SBJ you) (VP know))) (PRN , (S (NP-SBJ I) (VP mean)) ,) (NP-SBJ you) (VP hear (PP about (S-NOM (NP-SBJ people) (VP getting (VP mugged)))))))) . E_S)) Final punctuation: the rule that commas go inside PRN brackets is overridden by the final punctuation rule, that says that final punctuation goes at top level. ( (S (NP-SBJ I) (VP-UNF am (PRN , (S (NP-SBJ you) (VP know)))) , N_S)) (RS ]) should be treated like punctuation. This means that if the final punctuation is followed by (RS ]), then both of them go at top level. ( (S-UNF (INTJ like) (EDITED (RM [) (NP-SBJ I) , (IP +)) (RS ]) (SBAR-TMP (WHADVP-1 when) (S (EDITED (RM [) (S (NP-SBJ I) (VP was (NP-PRD-UNF h-))) , (IP +)) (NP-SBJ I) (VP went (ADVP-DIR home) (ADVP-TMP *T*-1)))) , (RS ]) N_S)) Elsewhere (RS ]) goes at the highest level available, just like other punctuation. ( (SBARQ (WHADJP-1 How (EDITED (RM [) s- , (IP +)) deprived) (RS ]) (SQ could (NP-SBJ they) (VP be (ADJP-PRD *T*-1) (SBAR-ADV if (S (NP-SBJ they) (VP had (NP a camcorder)))))) ? E_S)) In restarts there should really only be one constituent (usually UNF). If there are INTJs hanging around put them inside something. ( (SQ And (EDITED (RM [) (EDITED (RM [) (SQ-UNF were (NP-SBJ they)) , (IP +)) (SQ-UNF (INTJ uh) , were (NP-SBJ they)) , (RS ]) (IP +)) were (NP-SBJ they) (RS ]) (ADVP obviously) (ADJP-PRD-UNF poor or depri- (PRN , (S (NP-SBJ you) (VP know)))) ? E_S)) Stray 'uh's and other INTJ type stuff There is a diffle rule that says to put 'Uh' at the end of the previous sentence (when ending with a period) IF nothing follows the 'Uh'. Sometimes difflers do this even when something follows. These need to be split off and added to what follows. In the same vein, they sometimes put 'well', 'I mean', and 'so' at the end of sentences. These should ALWAYS be split off from the ends of sentences, whether or not anything follows (if it doesn't they can be S-UNF by themselves). ( (S (NP-SBJ I) (VP do not (VP know (SBAR (WHNP-1 what) (S (NP-SBJ the world) (VP (VP is (NP-PRD *T*-1)) or (VP is not (NP-PRD *T*-1))))))) .)) ( (S (INTJ Uh) , E_S (NP-SBJ-UNF the dimi-) , N_S)) =============================================================================== THE...THE construction The first THE thing is the head of a sort of comparative (which I've jazzed up a bit by doing it properly with an empty operator and trace, as comparatives should be), while the second THE is a topic. ( (S (ADVP-MNR (ADVP the faster) (SBAR (WHADVP-1 0) (S (NP-SBJ you) (VP eat (ADVP-MNR *T*-1))))) (ADJP-PRD-TPC-2 the sicker) (NP-SBJ you) (VP get (ADJP-PRD *T*-2)))) Since the first THE thing is just floating around in the sentence it has to be adverbial, which is fine if the THE thing is an adverb. If it is an NP, then it can be NP-ADV. ( (S (NP-ADV (NP the more money) (SBAR (WHNP-1 0) (S (NP-SBJ you) (VP make (NP *T*-1))))) (NP-TPC-2 the more books) (NP-SBJ you) (VP can (VP buy (NP *T*-2))))) Often though it is ADJP, in which case I guess we have to do it like those other floating adjectival things, with an S-ADV and empty subject (or is there a better way?). ( (S (S-ADV (NP-SBJ *) (ADJP-PRD (ADJP the better fed) (SBAR (WHADJP-1 0) (S (NP-SBJ the lion) (VP is (ADJP-PRD *T*-1)))))) (ADJP-PRD-TPC-2 the safer) (NP-SBJ the trainer) (VP is (ADJP-PRD *T*-2)))) It's not uncommon for there to be inversion, in which case the whole thing will be SINV ( (SINV (S-ADV (NP-SBJ *) (ADJP-PRD (ADJP the better fed) (SBAR (WHADJP-1 0) (S (NP-SBJ the lion) (VP is (ADJP-PRD *T*-1)))))) (ADJP-PRD-TPC-2 the safer) (VP is (ADJP-PRD *T*-2)) (NP-SBJ the trainer))) And often in one or the other part (or both) the verb is elided. For this use (VP *?*) and put the trace inside it. ( (S (S-ADV (NP-SBJ *) (ADJP-PRD (ADJP the better fed) (SBAR (WHADJP-1 0) (S (NP-SBJ the lion) (VP *?* (ADJP-PRD *T*-1)))))) (ADJP-PRD-TPC-2 the safer) (NP-SBJ the trainer) (VP *?* (ADJP-PRD *T*-2)))) If it's really reduced, like 'the more, the merrier', then just do it as FRAG. ( (FRAG (NP/ADVP? the more) (ADJP the merrier))) Here are some real examples culled from the old lists and mail messages. ( (S (ADVP-TMP (ADVP The sooner) (SBAR (WHADVP-1 0) (S (NP-SBJ you) (VP act (PP-MNR like (NP an angel)) (ADVP-TMP *T*-1))))) (ADVP-TPC-2 the quicker) (NP-SBJ you) (VP 'll (VP feel (ADJP-PRD angelic) (ADVP *T*-2))) )) ( (S (ADVP-MNR (ADVP The tighter) (SBAR (WHADVP-1 0) (S (NP-SBJ you) (VP squeeze (ADVP-MNR *T*-1))))) , (ADVP (ADVP the more) (SBAR (WHADVP-2 0) (S (NP-SBJ the price) (VP goes (PRT up) (ADVP *T*-2))))) , (NP-PRD-TPC-3 the more incentive) (NP-SBJ there) (VP is (NP-PRD *T*-3)) . E_S)) ( (S (PP of (NP course)) , (ADVP-TMP (ADVP the longer) (SBAR (WHADVP-1 0) (S (NP-SBJ it) (VP did not (VP happen (ADVP-TMP *T*-1)))))) , (ADJP-PRD-TPC-2 the stronger) (NP-SBJ her wish and belief (SBAR that (S (NP-SBJ it) (VP might not (VP *?*))))) (VP *?* (ADJP-PRD-2 *T*-2)))) ( (SINV (S-ADV (NP-SBJ *) (ADJP (ADJP the more extensive) (SBAR (WHADJP-1 0) (S (NP-SBJ the credo) (VP *?* (ADJP-PRD *T*-1)))))) (ADJP-PRD-TPC-2 the more unified and strong) (VP is (ADJP-PRD *T*-2)) (NP-SBJ the group))) ( (SINV (S-ADV (NP-SBJ *) (ADJP-PRD (ADJP The blunter) (SBAR (WHADJP-1 0) (S (NP-SBJ the knife) (VP *?* (ADJP-PRD *T*-1)))))) , (ADJP-PRD-TPC-2 the higher) (VP is (ADJP-PRD *T*-2)) (NP-SBJ (NP the value) (PP for (NP A[fj]))))) ( (S (S-ADV (NP-SBJ *) (ADJP-PRD (ADJP The longer) (SBAR (WHADJP-1 0) (S (NP-SBJ your distribution list) (VP *?* (ADJP-PRD *T*-1)))))) , (NP-TPC-2 the more) (NP-SBJ-3 you) (VP save (NP *T*-2) (S-ADV (NP-SBJ *-3) (VP using (NP the list processor)))))) ( (S (S-ADV (NP-SBJ *) (ADJP-PRD (ADJP the more extensive and firm) (SBAR (WHADJP-1 0) (S (NP-SBJ the body of doctrine) (VP *?* (ADJP-PRD *T*-1)))))) , (ADJP-PRD-TPC-1 the firmer) (NP-SBJ the group) (VP *?* (ADJP-PRD *T*-1)))) ( (S And (S-ADV (NP-SBJ *) (ADJP-PRD (ADJP the more complex) (SBAR (WHADJP-1 0) (S (NP-SBJ the morphophonemic system) (VP is (ADJP-PRD *T*-1) (PP in (NP (NP relation) (PP to (NP the phonemic base))))))))) , (ADVP-MNR-TPC-2 the less easily) (NP-SBJ-3 a phonemic system) (VP will (VP be (VP analysed (NP *-3) (ADVP-MNR *T*-2) (PP without (NP (NP close attention) (PP to (NP the morphophonemics))))))))) ( (S (ADVP-TMP (ADVP The sooner) (SBAR (WHADVP-1 0) (S (NP-SBJ you) (VP act (PP-MNR like (NP an angel)) (ADVP-TMP *T*-1))))) (ADVP-TPC-2 the quicker) (NP-SBJ you) (VP 'll (VP feel (ADJP-PRD angelic) (ADVP *T*-2))))) ( (SINV (ADVP (ADVP the more) (SBAR (WHADVP-1 0) (S (NP-SBJ-2 the action) (VP claims (S (NP-SBJ *-2) (VP to (VP be (ADJP-PRD total) (ADVP *T*-1)))))))) , (ADJP-PRD-TPC-3 the smaller) (VP is (ADJP-PRD-3 *T*-3)) (NP-SBJ (NP (NP the part) (PP of (NP man))) (VP engaged)))) ( (S (NP-SBJ it) (VP follows (SBAR that (S (S-ADV (NP-SBJ *) (ADJP-PRD (ADJP the more eminent) (SBAR (WHADJP-1 0) (S (NP-SBJ the victim) (VP * (ADJP-PRD *T*-1)))))) , (ADJP-PRD-TPC-2 the more impressive) (NP-SBJ the lesson) (VP *?* (ADJP-PRD *T*-2))))))) ( (S (NP-ADV (NP The more factories and robots) (SBAR (WHNP-1 0) (S (NP-SBJ Japanese manufacturers) (VP add (NP *T*-1))))) , (S (NP-TPC-2 the more) (NP-SBJ-1 they) (VP will (VP be (ADJP-PRD able (S (NP-SBJ *-1) (VP to (VP export (NP *T*-2)))))))) , and (S (NP-TPC-3 the less) (NP-SBJ-4 their domestic customers) (VP will (VP need (S (NP-SBJ *-4) (VP to (VP import (NP *T*-3))))))))) =============================================================================== Below is the typo policy we developed for the POS taggers. Since there seems to be a fair amount of variation in using the TYPO label in bracketing, we're going to adopt the same policy. Basically, use a TYPO label in any of the cases listed below where the POS taggers are supposed to use the typo sign ^. In the cases where the tag /GW is used, enclose all the parts which are supposed to be tagged /GW as well as the word actually tagged with the typo sign. I've put the relevant parts of the policy for /GW use at the end. Please read this carefully because it outlines the cases where you can and cannot assume a hyphen (i.e., the use of a GW tag implies a hyphen), and sets up "noah" (the electronic Webster's) as the final arbitrator of whether or not the collocation is hyphenated. =============================================================================== Switchboard Typo Policy Typos are indicated by a caret (^) preceding the tag. The tag given is the tag for the hypothesized correct word, not the actual word. Words which are wrongly capitalized (or wrongly not capitalized) should not on that account alone count as typos. They should be tagged as nouns or proper nouns according to sense not capitalization. Proper names which are wrongly spelled may be labelled typos but it is not necessary. The typos exemplified here are common and fairly certain. There are many other cases which are more shaky. As a general rule, something should only count as a typo if it is a homophone or a near homophone (defined as not more than one sound (vowel or consonant) different), but you'll have to use your judgement about individual cases. There are some examples of things that are definitely not typos at the end of this message. Feel free to ask for a second opinion if you're in doubt. A.1 Homophones When the typo is a homophone of the correct word it is tagged with the typo sign (^) and the tag for the correct word. put/VB there/^PRP$ money/NN in/IN places/NNS (=their) it/PRP 's/BES really/RB not/RB to/^RB bad/JJ (=too) their/PRP$ reporters/NNS needed/VBN to/^CD shots/NNS (=two) know/^DT matter where you build it, (=no) right/^VB a book about it (=write) I/PRP get/VBP to/TO here/^VB about/IN Texas/NNP (=hear) fate through/^VBD them together / (=threw) you/PRP could/MD say/VB high/^UH to/TO your/PRP$ teachers/NNS (=hi) He one/^VBD the race (=won) I/PRP might/MD of/^VB ./. (=have) A.2 Semi-homophones (or homophones in some dialects) Some words that are not homophones in standard English are in speech or at least in some dialects. Some of these are pretty standard (like final t/d deletion) while others (higher=here) are heavy dialect. Some common examples are listed below. (a) final t/d deletion everybody/NN is/NNS suppose/^VBN to/TO bring/VB something/NN I/PRP use/^VBD to/TO play/VB racquetball/NN --/: I/PRP would/MD have/VB like/^VBN to/TO have/VB ham and bake/^VBN potatoes, (b) an/and confusion An/^CC they/PRP have/VBP to/TO be/VB in/IN ideal/JJ physical/JJ shape/NN ,/, basically/RB ./. (an=and) there/EX 's/BES and/^DT old/JJ joke/NN about/IN (and=an) (c) than/then confusion other/JJ then/^IN that/IN (then=than) more reluctant of letting my older children go baby-sit for her because I didn't know her -- --/: then/^IN she/PRP was/VBD reluctant/JJ of/IN letting/VBG strangers/NNS into/IN her/PRP$ house/NN (then=than) (d) where/were/we're the/DT fact/NN that/IN people/NNS where/^VBD having/VBG this/DT problem/NN when/WRB you/PRP where/^VBD in/IN school/NN there/EX where/^VBD large/JJ numbers/NNS not were/^WRB we are. and/CC were/^PRP^VBP officially/RB ,/, yeah/UH ,/, and/CC we/PRP 're/VBP officially/RB in/IN a/DT state/NN of/IN emergency/NN ./. And/CC were/^PRP^VBP here/RB (e) I/a/uh as/IN a/^PRP say/NN (a=I) where/WRB a/^PRP do/NN n't/RB (a=I) it/PRP takes/VBZ almost/RB six/CD months/NNS to/TO get/VB ,/, uh/^DT ,/, handgun/NN permit/NN (uh=a) 0926 I/PRP think/VBP that/DT 's/BES I/^DT pretty/RB good/JJ idea/NN ./. (I=a) (f) accept/except, access/excess Accept/^IN for what we give to my daughter I think other then/^IN accept/^IN on a commercial or on news coverage (note also 'then' for 'than') in/IN access/^NN of/IN that/DT now/RB so/IN it/PRP (g) are/or Are/^CC they put one on the parents the/DT first/JJ four/CD ,/, five/CD years/NNS or/^VBP so/RB important/JJ just because people or/^VBP so, I don't know, just today people are just so money hungry, (h) i/e confusion they/PRP did/VBD n't/RB have/VB a/DT since/^NN of/IN risk/NN (since=sense) I well/^MD never, ever go back. (well=will) had/VBD some/DT bends/^NNS (from context clearly = bins) (i) s/c confusion investment advise/^NN somebody/NN trying/VBG to/TO ,/, to/TO device/^VB a/DT scam/NN (i) miscellaneous (some of these are probably keyboard mistakes) I/PRP 'd/MD a/^VB never/RB put/VBN (a=have) Here/^PRP$ name/NN is/VBZ Lori/NNP (here=her) no/DT skin/NN of/^IN my/PRP$ back/RB (of=off) hugh/^JJ companies like Three M, uh, I'm, uh, Honeywell (hugh=huge) our/PRP$ cancer/NN society/NN sales/^VBZ daffodils/NNS ./. (sales=sells) I think it's four cups of floor/^NN , (floor=flour) the/DT okra/NN that/WDT is/VBZ growing/VBG around/IN higher/JJR (higher=here?) A.3 Keyboard mistakes If a word has one wrong letter or is missing a letter and it is obvious from context what the word should be, count it as a typo. The following sets at least are allowable as typos. (a) than/that he's a better man that/^IN I am (that=than) (b) if/it/is/in in/^PRP was/VBD in/IN high/JJ school/NN (in=it) somebody does if/^PRP and breaks tradition (if=it) especially it/^IN he's a repeater (it=if) Is/VBZ is/^PRP just/RB aerobics/NN or/CC ,/, (is=it (probably)) (c) on/of/or take it out or your paycheck (or=of) (d) to/do/so So/RB ,/, what/WP else/RB to/^VBP you/PRP tape/VBP (to=do) What/WDT sort/NN of/IN requirements/NNS to/^VBP you/PRP have/VBP (to=do) just/RB so/^TO see/VB what/WP ,/, uh/UH ,/, they/PRP might/MD have/VB to/TO offer/VB (so=to) push/VB it/PRP so/^IN the/DT end/NN --/: (so=to) (e) out/our great/JJ out/IN country/NN really/RB is/VBZ ./. (out=our) (f) the/they that/DT 's/BES what/WP the/^PRP prefer/VBP to/TO do/VB (the=they) they/^DT one we was all worried about -- (they=the) (g) miscellaneous (some might be dialect rather than keyboard mistakes) I/PRP do/VBP n't/RB thing/^VB they/PRP (thing=think) just/RB short/^RB of/IN have/VBP it/PRP on/IN (short=sort) does/VBZ [ you/^PRP$ husband/NN ] (you=your) I/PRP was/VBD just/RB wandering/VBG if/IN (wandering=wondering) he's got now/^DT doubt that (now=no) one/CD or/CC two/CD timer/^NNS a/DT year/NN (timer=times) whether/IN I/PRP night/^MD ,/, uh/UH ,/, uh/UH ,/, (night=might) we/PRP 're/VBP taking/^VBG about/IN monies/NNS way/RB (taking=talking) B. Words that should be split but aren't A word that should be split is given two tags, one for each of the parts. Each of the tags is preceded by the typo sign (^) (a) cliticized verbs not separated If/IN your/^PRP^VBP happy/JJ with/IN it/PRP (=you 're) their/^PRP^VBP offering/VBG a/DT service/NN (=they 're) whose/^WP^BES going/VBG to/TO really/RB make/VB them/PRP ./. (=who 's (who is)) Someone whose/^WP^HVS got accounts (=who 's (who has)) Okay/UH ,/, I/PRP guess/VBP its/^PRP^BES recording/NN ./. (=it 's) my/PRP$ husbands/^NN^BES retired/VBN ./. (=husband 's) [ The, + the ] deposits/^NN^BES only on like drink stuff. (=deposit 's) who/WP do/VBP you/PRP thinks/^VB^BES going/VBG to/TO win/VB (=think 's) (b) cliticized pronoun not separated lets/^VB^PRP turn/NN the/DT war/NN off/RP (=let 's) (c) singular possessive /'s/ not separated what/WP the/DT guys/^NN^POS name/VBP is/VBZ (=guy 's) my/PRP$ in-laws/^NN^POS place/NN (=in-law 's) my/PRP$ neighbors/^NN^POS yard/NN ./. (=neighbor 's) (d) plural possessive apostraphe missing (if it were present it would be tagged POS) your cats/^NNS^POS names (=cats ') the students/^NNS^POS , uh, parents (=students ') (e) verb-particle combinations (but NOT noun-particle combinations, which should be joined, and must be joined up using GW if they are separated) people/NNS giveaway/^VBP^RB personal/NN information/NN (=give away) It/PRP was/VBD already/RB setup/^VBN^PRT (=set up) If he doesn't back-up/^VB^PRT (=back up) (f) other miscellaneous cases for/IN along/^DT^JJ time/NN (=a long) for/IN awhile/^DT^NN ./. (=a while) want/VB to/TO have/VB anymore/^DT^JJ children/NNS (=any more) that/DT maybe/^MD^VB true/JJ (=may be) we went fishing everyday/^DT^NN (=every day) C. Words that are separate but shouldn't be Parts of words that are separated are joined with the GW tag. The final part carries the label for all the joined parts (not usually more than two) and it has the typo sign preceding its label. Note that the typo sign is used on the final part even if it would have the right tag anyway (see the 'with out' example below). (a) plural /s/ written as possessive other/JJ than/IN just/RB it/GW 's/^PRP$ labor/NN (=its) the/DT Honda/GW 's/^NNPS have/VBP been/VBN very/RB safe/JJ ./. (=Hondas) I wouldn't be surprised if thing/GW 's/^NNS like that didn't happen (=things) our/PRP$ causes/NNS do/VBP not/RB seem/VB as/RB important/JJ as/IN you/PRP all/GW 's/^PRP were/VBD (=you alls) ?? (b) third singular /s/ written as possessive Well/UH ,/, what/WP get/GW 's/^VBZ me/PRP is/VBZ (=gets) (c) they're = their, you're = your they/GW 're/^PRP$ Mom would never know it. (=their) I gave him you/GW 're/^PRP$ book (=your) (b) 'a-' words well/UH ,/, let/VB me/PRP go/VB a/GW head/^RB (=ahead) they don't look a thing a/GW like/^JJ . (=alike) There's a certain a/GW mount/^NN of dribble (=amount) (c) 'all-' words all/GW though/^IN I have a feeling that people look (=although) It/PRP 's/BES all/DT metric/JJ all/GW ready/^RB ./. (=already) (d) miscellaneous examples a/DT meal/NN with/GW out/^IN any/DT vegetables/NNS (=without) I don't like him any/GW more/RB (=anymore) D. Definitely not typos (a) numbers in words indicate accents, just ignore them ,/, and/CC my/PRP$ husband/NN ,/, then/RB fianc3e/NN (b) don't correct people's grammar, if they use the wrong tense of a verb or whatever, just leave it latest/JJS one/NN I/PRP 've/VBP saw/VBD (c) 'zero' spelled 'oh' is a CD and is not a typo the/DT first/JJ appearance/NN of/IN Roger/NNP Moore/NNP as/IN double/JJ oh/CD seven/CD we/PRP actually/RB do/VBP have/VB some/DT money/NN in/IN a/DT Four/CD Oh/CD One/CD K/SYM BEVERLY/NNP HILLS/NNP NINE/CD OH/CD TWO/CD ONE/CD OH/CD ,/, =============================================================================== New Policy for use of GW GW is to be used only for correcting typos which affect tokenization of words; that is, when the transcriber spelled as two separate words something that should be spelled as one word. There are some very clear cases, like 'a like' for 'alike', 'back ground' for 'background', and many less clear cases. The final arbitrator for whether two items go together to make one word will be the dictionary, Noah. If you look the collocation up in Noah and it gives it as one entry spelled with only hyphens between the parts then you can use GW. So if you look well rounded up in Noah it gives the following: [well-round-ed] (we_l'roun'di_d) (ADJECTIVE). 1. Comprehensively developed: ``a well-rounded scholar.'' 2. Having a shapely figure. Since the parts are separated only by hyphens, this is a possible place to use GW. If Noah does not reckonize the collocation, or recognizes it but spells it with spaces rather than hyphens between the parts, this indicates that GW is NOT possible. Thus, while 'well rounded' appears in Noah as 'well-rounded' and can therefore be tagged 'well/GW rounded/JJ', 'well dressed' is not recognized, so must be tagged 'well/RB dressed/JJ'. It is not, however, necessary to tag hyphenated collocations with GW, so 'well rounded' can also be tagged 'well/RB rounded/JJ'. In general, try to use GW only when it is not possible to acurrately tag the two parts separately. Places where GW should be used separated prefixes pre- pre/GW med/JJ student/NN pre/GW regime/NN crimes/NNS mini- mini/GW series/NNS mini/GW skirt/NN But note that when standing alone as a short form for 'miniskirt' (she wore a mini/NN), it is a noun inter- inter/GW state/NN inter/GW United/NNP States/NNPS ex- ex/GW husband/NN ex/GW boy/NN friend/NN But note that 'my ex' would be my/PRP\$ ex/NN semi- the semi/GW final/JJ game/NN a semi/GW nice/JJ dinner/NN But note that 'semi', like 'mini', can also be used as a noun, i.e., a type of truck (he drives a semi/NN) non- non/GW fiction/NN books/NNS Nouns derived from verb-particle collocations are considered single words and thus when separated should be joined with GW the kids are drop/GW outs/NNS he's a real cut/GW up/NN we had a break/GW down/NN all kinds of nice write/GW ups/NNS about it did you do your work/GW out/NN this morning? a sixty/CD five/CD percent/NN turn/GW out/NN the reason for the split/GW up/NN BUT NOTE THAT THIS IS ONLY THE CASE WHEN THESE COLLOCATIONS ARE **NOUNS**. ADJECTIVAL AND PASSIVE USE OF THESE COLLOCATIONS ARE TAGGED SEPARATELY. SEE BELOW. Two word verbs should be joined by GW. Do you tent/GW camp/VB or do you have a camper If your child back/GW talks/VBP I substitute/GW teach/VBP I always over/GW pay my deductions You can easily over/GW do/VB it He always dry/GW cleans/VBP his shirts (BUT NOTE: dry/JJ cleaner/NN, dry/JJ cleaning/NN) They drug/GW test/VBP and it's not random (BUT NOTE: they gave him a drug/NN test/NN; they do drug/NN testing/NN ) I was trying to saltwater/GW fish/VB (BUT NOTE: I went surf/NN fishing/NN ???????? The dog was being house/GW sit/VBN They were down/GW grading/VBG all these other ones Places NOT to use GW and what to do instead In general it won't be correct to join together prepositions with GW. Although 'into' and 'onto' are indeed different from 'in to' and 'on to', in most cases both will be possible in the same contexts. The difference between 'he put it into the bag' and 'he put it in to the bag' is only one of stress, which we, of course, don't have access to. So unless it's absolutely impossible to read the sentence with a stress on each preposition, do not join them. I went in/IN to/IN a hardware store what would you be most interested in/IN getting in/IN to/IN are you rich or poor or in/IN between/IN he moved on/IN to/IN L A the ability to move on/IN to/IN some new technology Collocations with the ``suffix'' wise should be tagged separately with wise tagged as an adverb. you probably pay more percentage/NN wise/RB he's doing well popularity/NN wise/RB age/NN wise/RB he's at the upper limit Collocations with 'type' should also be done as separate words a dinner/NN type/NN thing/NN a polluting/NN type/NN deal/NN a career/NN type/NN position/NN P/NN C/NN type/NN things/NNS sight/NN seeing/NN type/NN stuff/NN an Austin/NNP Heely/NNP type/NN engine/NN Collocations with 'free' should be done as separate words lint/NN free/JJ violence/NN free/JJ Nouns derived from verb-particle collocations (dropout, breakdown, etc.) are considered single words (see above). But when used adjectivally or in the passive, these are tagged separately. the cut/VBN up/RP tomatos/NNS a built/VBN in/RP window/NN box/NN this beaten/VBN down/RP path/NN they even have dress/VB up/RP days/NNS the teachers I know wear dress/VB up/RP jeans/NNS a call/VB in/RP survey/NN a fix/VB up/RP special/NN pull/VB out/RP couches/NNS it had some well/RB thought/VBN out/RP parts/NN watered/VBN down/RP wine/NN he got himself really messed/JJ up/RP he just got fed/JJ up/RP I'm just kind of spaced/JJ out/RP we're so wound/JJ up/RP in the boy scouts I just feel hemmed/JJ in/RP by that everyone is spread/JJ out/RP all over TImbuktu I don't typically feel intruded/JJ on/RP their home is paid/JJ for/RP The pages were torn/VBN out/RP by my little sister Multi-word modifiers of any type should be tagged word by word (unless Noah gives them as hyphenated, in which case you MAY, but don't have to use GW) well/RB dressed/VBN politicians defense/NN oriented/VBN military/NN agression/NN oriented/VBN militaryNN a state/NN sponsored/VBN school/NN fast/RB breaking/VBG events/NNS the eye/NN stinging/VBG variety/NN time/NN consuming/VBG projects/NNS problem/NN solving/VBG skills/NNS the wood is very slow/RB burning/JJ sloppy/JJ looking/JJ jeans/NNS funny/JJ looking/JJ coke/NN strange/JJ looking/JJ can/NN country/NN looking/JJ watermelon/NN blond/JJ headed/JJ girl/NN open/JJ minded/JJ person/NN liberal/JJ minded/JJ politician/NN the/DT lower/JJR end/NN of/IN the/DT top/NN of/IN the/DT line/NN hotels/NNS break/NN for/IN out/IN of/IN state/NN students/NNS an/DT up/RB and/CC coming/VBG team/NN out/IN of/IN body/NN experiences/NNS he was on a/DT year/NN and/CC a/DT half/JJ training/NN plan/NN