When we say that each 'word' is assigned a tag, we mean each graphic word, i.e. each sequence of characters surrounded by spaces (or punctuation marks).10 This leads to differences in tagging where there is variation in spelling, as in:
7 Problem areas
The aim of the sections below is to show as explicitly as possible how a variety of problems were dealt with in the tagged LOB Corpus. Though the focus is on explaining actual practice in the corpus, there are also some remarks on what might have been done otherwise. In hindsight, it is easy to see that some things could have been done differently. This is a common enough experience in a large project; it is only at the end that one knows how to carry it out. We hope the sections below will be of interest both to the user of the tagged LOB Corpus and to those who embark on similar projects in the future.The examples used in the discussion below are, with few exceptions, authentic corpus examples (whether references are given or not).9
7.1 Word division
When we say that each 'word' is assigned a tag, we mean each graphic word, i.e. each sequence of characters surrounded by spaces (or punctuation marks).10 This leads to differences in tagging where there is variation in spelling, as in:
|
fancy free |
NN JJ |
science fiction |
NN NN |
|
fancy-free |
JJ |
science-fiction |
NN |
|
|
|
|
|
|
first rate (works) |
OD NN |
every day (life) |
AT NN |
|
first-rate (works) |
JJ |
everyday (life) |
JJB |
Another consequence is that we get some very strange 'words', in particular hyphenated sequences, as in: an above-the-rooftops position. Most of these, however, are found only in attributive position and are given the tag JJB (see 7.8) and are therefore identified as more or less idiosyncratic. More seriously, the graphic word division sometimes conflicts sharply with syntactic boundaries, as in:
|
Henry |
8's hall |
the New |
York-born Negress |
|||
|
|
|
|||||
|
a £6 |
10 |
\0s-a-week |
cook |
Spanish and |
Portuguese-speaking |
|
In some cases of discrepancy of this kind we deviated from our 'one-word-one-tag' approach and resorted to idiom tagging, particularly in the case of recurrent multiword sequences functioning as units.
7.2 Idiom tagging
By idiom tagging we understand the treatment of a sequence as a unit using ditto tags (cf 3.3 and 4.4). Ditto tags were introduced at a late stage of the project and were used rather sparingly. Their use is related to: (1) the frequency of the sequence, (2) the degree of unacceptability or arbitrariness of single-word tagging, and (3) the chances of unambiguous identification of the sequence. The 'idiom list' through which the ditto tags were assigned contained a fairly limited number of high-frequency combinations. Sequences like so as to (TO TO"TO" rather than CS CS TO) and one another (PPLS PPLS" rather than CD1 DT) qualified according to all three criteria. A combination like at large qualifies according to the second (and perhaps also the first) criterion but failed the crucial third criterion and was therefore not included in the 'idiom list'. Another sequence not included was by and large, which satisfies the last two criteria. In a good number of cases (including the last two examples) ditto tag, s were introduced during post-editing to avoid an unacceptable or arbitrary choice of tags at the level of the single word.
The principal cases of idiom tagging found in the tagged LOB Corpus are:
|
AP AP" (AP") |
a few, a little, a good many, a great many |
|
CC CC" CC" |
as well as |
|
CS CS" (CS") (CS") |
but that, considering that, except that, in order that, in that, now that, provided that, save that, seeing that, so that, such that, in as much as, inasmuch as, in so far as, insofar as, so as, such as, as if, as though, even if, even though |
|
IN IN"' (IN") |
according to, a la, along with, apart from, as against, as between, as for, as from, as of, as opposed to, as regards, as to, as versus, as well as, because of, but for, by means of, except for, excepting for, in front of, in spite of, inspite of, irrespective of, on board, on top of, owing to, prior to, save for, subsequent to, such as, thanks to |
|
JJ JJ" |
a priori, clear cut, de luxe, far off, fed up, hard up, old fashioned, so called, short lived, wilful appearing |
|
NN NN" |
alter ego, avant garde, carte blanche, in between, post mortem, rigor mortis |
|
NNS NNS" |
grown ups |
|
NNU NNU" |
per cent, per \0cent |
|
NPT NPT" |
\0M \0P |
|
PN PN" |
no one |
|
PN$ PN$" |
no one's |
|
PPLS PPLS" |
each other('s), one another('s)11 |
|
RB RB" (RB") |
a little, a priori, a posteriori, all but, all ready, all right, all round, anything but, and so on, and so forth, as good as, as well, as it were, as yet, at best, at first, at large, at last, at long last, at least, at length, at once, at random, at worst, by and by, by far, by and large, by(e) the by(e), far from, for certain, for ever, for good, for once, for sure, half way, in brief, in common, in full, in general, in particular, in private, in public, in short, in the main, in vain, inter alia, kind of, let alone, may be, more than, none the less, of a sudden, of course, of late, once more, per se, and again, vice versa |
|
TO TO" TO" |
in order to, so as to |
|
VB VB" |
button stitch, soft pedal |
There is a problem when these sequences, as occasionally happens, are interrupted: a good few, in order not to, so as not to, except perhaps for, apart altogether from; owing, however, to. As such interruptions are not very frequent, it was not considered necessary to devise a special mechanism for handling them.
Users of the tagged corpus will probably find that the use of idiom tagging is not quite consistent, and they may well wonder why certain sequences have not qualified. The reason is partly conflict between the three criteria mentioned above and partly the fact that the 'idiom list' was a rather late addition to the automatic tagging system and only minor adjustments could be made at the stage of post-editing.
For some comments on idiom tagging with different categories of words, see 7.10 (adverbs), 7.12 (determiners), 7.13 (prepositions), 7.14 (conjunctions), and 7.21 (foreign expressions).
7.3 -ed forms
An -ed form can be VBD, VBN, or JJ. There are also rare cases of IN (7.13, note 29), CS (7.14), and RB (7.18, the end). The choice of VBD vs VBN is occasionally problematic:
Should tested in (1) and the first occurrence of heard in (2) be interpreted as past-tense forms or present-perfect forms? In the former case we get questions with subject-verb order, in the latter ordinary interrogative structures with ellipsis of the initial auxiliary. Similarly, the second occurrence of heard in (2) and the verb forms in the second sentence of (3) could be regarded as past-tense forms or as present-perfect forms with ellipsis of the auxiliary. Apart from a small number of examples of this kind (dealt with according to the first principle in Section 6), the choice of VBD vs VBN is straightforward.
The major problem with -ed forms is the choice of VBN vs JJ. VBN is used when the -ed form is part of a complex verbal group (perfective aspect or passive) and when it occurs alone in non-finite clauses, as in:
The verbal character is shown by the type of complementation. Clear cases of JJ are:
1. bona fide adjectives ending in -ed: naked, wicked, etc.
2. -ed forms derived from nouns: talented, diseased, etc.
Since many words in English can be either nouns or verbs, it is not always easy to distinguish this type from verb forms, e.g. in:
Because of the coordination with a clear VBN form (decorated) and because of the following purpose clause, curtained is here probably best analysed as VBN.
3. prefixed or compound -ed forms like: unexpected, downtrodden, well-behaved, etc. The elimination of the ending does not result in an existing verb form: *unexpect, *well-behave, etc. However, forms beginning with un- may be problematic, as there are a good number of verbs with this prefix: undo, unlock, etc. Occasionally the same form can be analysed in two ways. Compare:
Because of the 'static' meaning in (7) and the 'dynamic' quality of the verbs in (8), the tags assigned were JJ vs VBN. See further the discussion below of problematic forms.
4. -ed forms as premodifiers of nouns: his injured leg, his well balanced compositions, the well established club, etc. The tag JJ is assigned even when the -ed form clearly has verbal characteristics (e.g. is preceded by an adverb which typically accompanies a verb) as in: admitably written books (F10:60), a badly curtained window (L17:67). Note the difference in tagging of pre- and postmodifying -ed forms (JJ vs VBN): the observed data (J32:55) vs the headfirst downward swimming observed in the field (J06:92); the above mentioned bodies (B10:215) vs the work mentioned in Chapter 17 (G02:62); etc.
5. -ed forms in nominal positions: the aged and the maimed, from the known to the unknowable, the seen and the unseen, the wounded, the most defeated of the inhabitants, etc. Since JJ is retained for clear adjectives in these cases (generic singular and plural forms and cases where a head noun is recoverable from the context; see 7.9), it should also be used for comparable -ed forms. Very rarely the reference is to an individual and a head noun is not recoverable from the context. The forms are then more noun-like and may even take genitive endings, which points towards NN: the accused, the deceased, her firstborn. Nevertheless, JJ was kept in these cases, since plural inflection does not occur (the other two accused A43:21). But note: coloureds NNS (A28:61, D17:62), newly-weds NNS (C04:55).
The principal problems arise with -ed forms after BE, which can either be a passive auxiliary (followed by VBN) or a copula (followed by JJ). VBN is used:
1. when there is a following by-phrase denoting an agent, as in: was surprised by ... ;
2. in other cases where the meaning is clearly 'dynamic' and a straightforward active equivalent is available: was built, was killed, etc;
3. with action verbs even when the meaning is 'stative'. Thus VBN is assigned in both of the following examples:
The structure of coordination in (10) shows that closed is similar to an adjective. Nevertheless, this form and other 'stative' -ed forms of action verbs have not achieved clear adjective status, as shown by their inability to accept intensification by very: *very closed, *very built, etc.
An -ed form after BE which cannot be identified as JJ on the basis of the criteria mentioned earlier is tagged JJ if two or more of the following conditions apply:
1. The form denotes a state (often an emotional state) rather than a process: agitated, confused, disappointed, surprised, etc.
2. The subject is human and denotes an experiencer. In contrast, the subject in a BE + VBN construction typically denotes a non-human entity affected by the action of the verb. Compare: she is disappointed vs the house is already sold.
3. BE can be replaced by the copula SEEM. Compare: she seems disappointed vs *the house seems already sold.
4. The -ed form can be preceded by the intensifier very. Compare: she is very disappointed vs *the house is very sold.
Forms tagged JJ often also have some of the following characteristics:
5. They can take prefixes characteristic of adjectives: inexperienced, overcrowded, uncomplicated, etc.
6. Coordination with clear adjectives is normal: happy and pleased, gentle and civilised, etc.
7. They can be replaced by synonyms or near-synonyms which are clear adjectives: tired, weary; complicated, complex; relaxed, calm; etc.
8. They are followed by prepositions which typically occur after adjectives. Compare: worried about, anxious about; pleased with, angry with; surprised at, angry at; etc. Note the difference in tagging depending upon the following preposition in cases like
|
surprised at |
JJ |
pleased with |
JJ |
worried about |
JJ |
|
surprised by |
VBN |
pleased by |
VBN |
worried by |
VBN |
9. The frequency of the -ed form in characteristically adjectival positions (attributive and predicative) is very high and outstrips the occurrences of clearly verbal uses.
Using the above criteria we can, for example, distinguish between:
Compare also:
JJ: |
... all her movements were limited and painful. F31:25 |
VBN: |
... the intake will still be strictly limited. B 18:95 |
In spite of all our criteria, we cannot claim that the tagging of -ed forms has been carried through consistently. The coordination criterion must be used with caution, as shown by the coordination of an adjective and a verb form in: Most marriages were polygonous and based upon a system of bride-price ... (J22:42-43). Sometimes the criteria conflict:
The prefixed -ed forms here should be JJ, since the elimination of the prefix does not result in an existing verb form. Nevertheless, the following by-phrase points in the direction of VBN. The former criterion was regarded as primary in this case, since it can be tested without inspection of context.
There definitely is no clear borderline between -ed adjectives and verb forms. In many cases we had to fall back upon the 'follow-the-tagger principle' (Section 6) and leave the ta chosen by the automatic tagging programs.
7.4 -ing forms
An -ing form can be V13G, JJ, NN, or RB (there are also rare cases of CS and IN; see 7.13 and 7.14). The RB tag is only applied rarely, in cases like: boiling hot, raving mad, screaming red, shining clean. There are three major classification problems with -ing forms: V13G vs JJ, V13G vs NN, and NN vs JJ.
VBG vs JJ
VBG is used when the -ing form is part of a complex verbal group (progressive aspect) and when it occurs alone in non-finite clauses, as in:
The verbal character is shown by the type of complementation. JJ rather than VBG is assigned:
1. where the elimination of the ending does not result in an existing verb form: awe-inspiring, deep-thinking, fierce - looking, hardworking, long-lasting, selfgoverning, thoroughgoing, uncompromising, etc.
2. where the -ing form is a premodifier in a noun phrase: an exciting performance; the existing pattern; a petty, tottering dictator; etc. JJ is assigned even when accompanying elements point toward VBG: the sound-producing apparatus (J32:7); power producing units (J78:148); with heavily beating heart (L04:91); etc. Note the difference in the tagging of pre- and postmodifying -ing forms (JJ vs VBG): a living man (J52:141) vs neither you nor any man living (P11:84); so many sitting targets (A23:98) vs a man sitting opposite her (N17:82); etc. The difference in tagging between the two positions reflects the degree of approximation (in structure and sequence) to an ordinary finite clause. Distinguishing between JJ and NN in premodifying position is often difficult; see below.
3. where the -ing form is used in a nominal position comparable to an adjective (ef 7.3 and 7.9), as in: the most promising of playwrights (C 14:205), spend it on the living (A36:110).
Difficulties in distinguishing VBG and JJ arise particularly after BE, which can either be a progressive auxiliary (followed by VBG) or a copula (followed by JJ).
The distinction is, however, clearer than that between VBN and JJ with -ed forms. VBG is assigned:
1. where there is a following object or another accompanying element characteristic of a verb, as in:
2. where the BE + -ing construction can be replaced by a single verb form without injuring the structure of the sentence. Compare: she is smiling, she smiles; it is raining, it rains; etc.
The tag JJ is assigned to -ing forms after BE which have the following characteristics:
1. They are related to transitive verbs which require an object or other post-verbal complement, but these are lacking and cannot be inserted after the -ing form. Compare: she surprises me vs she is surprising; it irritates me vs it is irritating; etc. Very often verbs of this kind take an experiencer object. An experiencer can be expressed by a prepositional phrase following the -ing form, as in:
It is interesting that there are many adjectival -ing and -ed forms related to the same verb. Compare: it surprises me, I am surprised (at it), it is surprising; it disappoints me, I am disappointed (with it), it is disappointing: it bores me, I am bored (with it), it is boring; it frightens me, I am frightened (of it), it is frightening; etc. Since adjectival and verbal -ed forms are harder to distinguish (cf 7.3), JJ-tagging is less widely used in the tagged corpus with -ed forms than with -ing forms in these cases.
Adjectival -ing forms after BE are not only related to verbs taking an experiencer object: statutes are binding (J49:37), cinema is not so compelling as TV (J26:31), neither was daring nor held views beyond the rest of the people (K05:135), the person is in every way deserving of the medal (H24:84), Mr Baring was less enterprising (K22:32), the results so far achieved are promising (J78:158), the British section which is very revealing (J66:30), etc.
2. BE + -ing cannot be replaced by a simple verb form without producing an ungrammatical sentence. Compare: it is amazing vs *it amazes; it is comforting vs *it comforts; it is frightening vs *it frightens; it is awfully becoming vs *it awfully becomes; it is binding vs *it binds; etc.
3. BE can be replaced by the copula SEEM. Compare: she seems charming vs *she seems smiling.
4. The -ing form can be preceded by the intensifier very. Compare: she is very charming vs *she is very smiling.
Forms tagged JJ may also have the following characteristics:
5. They can often take prefixes characteristic of adjectives: unexciting. unconvincing, etc.
6. Coordination with clear adjectives is normal: long and boring; enthusiastic and stimulating:etc.
7. They can be replaced by synonyms or near-synonyms which are clear adjectives: surprising, remarkable; charming, delightful; etc.
8. They can often form adverbs by the addition of -ly: fittingly, strikingly, startlingly, surprisingly, etc.
9. The frequency of the -ing form in characteristically adjectival positions (attributive and predicative) is often high and outstrips the occurrences of clearly verbal uses.
Using the above criteria we can, for example, distinguish between the -ing forms in:
Three -ing forms after BE (lacking, missing, wanting) stand apart from the rest in lacking most of the adjectival characteristics. The tag D is used in examples like:
However, when there is a following object, the tag assigned is VBG:
Another special group consists of forms like: shining, beaming, glittering, sparkling. These are frequent in attributive position: shining eyes, sparkling wine, etc. Note the closeness to adjectives shown by the coordination in:
But clear adjectives can be coordinated with indisputable verb forms, as in:
20... a great shambling American poet big as a house and earning enough money to live in one in smart Belgravia... A19:198
Since the -ing forms in (16)-(17) fulfil one of our criteria for verb forms (cf above), the tag assigned was VBG.
VBG vs NN
Distinguishing VBG and NN is difficult with many -ing forms in nominal position. VBG is assigned to forms which are accompanied by objects or other elements characteristic of verbs, as in:
While genitive and possessive forms are inconclusive (see below), nominative or accusative 'subjects' identify an -ing form as VBG:
NN(S) is obviously used with -ing forms which have both the morphological and syntactic characteristics of nouns: building(s), painting(s), etc. Most often the grammatical status is revealed by syntactic features alone. NN is assigned to -ing forms preceded by determiners or modifiers characteristic of nouns, or followed by postmodifiers typical of nouns (especially of-phrases), as in:
Note that a nominal -ing form can be followed by an adverbial particle: needed a thorough shaking up (C04:102), the speeding up of the process (G65:148), the setting up of the Public Schools Commission (F28:112), etc. In such cases, however, the forms are usually hyphenated: building-up, rolling-up, winding-up, etc.
Sometimes there is a conflict between different clues in the context, as in examples like: 12
The determiner points towards NN, but since -ing forms in such constructions can take a direct object, which is perhaps our best criterion for verbal status, the tag assigned was VBG. Both noun- and verblike features are found with the coordinated -ing forms in
The coordinated -ing forms have here been treated differently, in accordance with the elements in the immediate context of each form.
Actual conflict between our criteria is not very frequent. More often difficulties arise where there is no sufficient indication in the context of VBG vs NN status. This applies to the following two cases
1. -ing forms preceded by possessive determiners and genitive form
Note that these can clearly be followed by VBG or NN, as shown by other elements in the context in:
Where there are no other clues in the context, the classification is problematic. The automatic tagging programs here tend to choose NN, which is no doubt often the correct choice, as in:
The verbs corresponding to the -ing forms in (41)-(44) require an object or other postverbal element, but these are lacking (in the third example the object relationship is expressed through the preceding possessive determiner). In (44) and (45) the superordinate verbs are of a kind which cannot be followed by a non-finite clause.
Where the possessive or genitive form expresses a subject relationship, a case can often be made for VBG, as in:
Here it is possible to insert clear verb constructions. Compare:
Many examples do not contain any clues like those mentioned above, e.g.:
In such cases the general rule was to leave the tag assigned by the automatic tagging programs (cf Section 6). Some of the tests taken up under the next point were, however, helpful.
2. Naked -ing forms
The most serious problem is handling 'naked' -ing forms in nominal positions. Where the elimination of the ending does not result in an existing verb form, we must clearly tag NN: book-making, ballet-composing, glass-blowing, stonethrowing, adventuring, etc. Note also that a 'naked' -ing form of a transitive verb is tagged NN, because of the lack of an object, as
Meaning is also a good guide here, especially where the -ing form is used in a concrete sense (as in 57 and 58). Coordination with a clear noun or noun phrase is a further clue (as in 58 and 59); the shared modifier in the last example is especially significant.
The analysis of 'naked' -ing forms of transitive verbs is less straightforward in object position after: deserve, need, require, want. Examples (the only instances found in the corpus occurred with NEED):
The meaning of the -ing form is equivalent to a passive to-infinitive. Compare:
An -ing form after NEED can clearly be NN, as shown by the determiner in:
Clear evidence of verbal status is hard to find. Compare:
In spite of this, the general rule has been to use the tag VBG for 'naked' -ing forms after NEED, provided (1) that they can be paraphrased by a passive to-infinitive and (2) that the object is recoverable from the subject of the superordinate clause.
The greatest difficulties arise with 'naked' -ing forms of intransitive verbs (or verbs which have both transitive and intransitive uses). Examples:
A 'naked' -ing form of an intransitive verb can clearly be either NN or VBG. Consider the following examples:
As shown by the possible additions (within parentheses), the forms can combine either with adverbs (as in 72 and 74) or adjectives (as in 73 and 75). In the former case there is a specific subject recoverable from the superordinate clause.
Where a specific subject is recoverable from the context, the tag assigned is V13G, as in:
If the -ing form occurs in a position which cannot be occupied by a clear noun or noun phrase, the tagging V13G would seem natural., e.g. after the intransitive verbs in:
But while there are parallels to (79) with clear verb constructions, we find examples like (80) with -ing forms which we would not normally classify as verbs. Compare:
The construction illustrated in (80) and (82) does, in fact, accept -ing forms which cannot easily be classified: go flat-hunting, go potholing, go sightseeing, etc. There are no verbs *pot hunt, *flat-hunt, *pothole, *sightsee. Nevertheless, where the GO + -ing construction is used, we may also find BE + -ing, which is a clear verb construction (progressive aspect): were pot hunting, were flat-hunting, were potholing, were sightseeing. Since these -ing forms can clearly occupy verbal positions, the best tag seems to be VBG.
A naked -ing form receives the tag NN under the following conditions (cf above):
1. where the elimination of the ending does not result in an existing verb form (but note the problem raised by examples like 82);
2. where the corresponding verb is transitive (but note the treatment of examples like 60 and 6 1).
In assigning NN we have also been guided by the following considerations:
3. If the form in -ing is quite established in a nominal function, NN is often the best tag: advertising, engineering, housing, training, etc. Such forms often have special meanings and are listed as separate entries in dictionaries.
4. If there are a good number of clear examples of a nominal -ing form elsewhere in the same text, NN may be extended to uncertain cases.
5. If a specific subject cannot be recovered from the context, NN may be preferable to VBG (cf 76-78 examples). Note, however, that there is no specific subject recoverable from the context with many clear cases of VBG, as in:
6. Coordination with clear nouns or noun phrases points towards NN. Examples (see also 58 and 59 above):
The evidence in (92) may seem inconclusive, but since the genitive form does not express a subject relationship (cf above), NN is the tag to be preferred and can be assigned to the three coordinated -ing forms. Bruising in (93) is related to a transitive verb and is therefore NN (cf above) and the same tag can be used for swelling. The following examples show the limitation of the coordination criterion:
In (94) we find coordination of a clear noun and a clear VBG construction. In (95) a prepositional phrase is coordinated with a problematic form like that in example (82) above and a 'naked' -ing form which can be expanded to a finite verbal group (whether he was fishing) and therefore must be VBG. Our 'leaking' criteria do not allow consistent classification of many 'naked' -ing forms. Often it was necessary to resort to the 'follow-the-tagger principle' (Section 6) and thus leave the tag assigned by the automatic tagging programs.
NN vs JJ
Note first that JJ is used when an -ing form occurs in a nominal position comparable to an adjective; cf above (VBG vs JJ). The main difficulty arises with -ing forms in attributive position before a noun. Here we can eliminate VBG, since verbs do not generally occupy this position and the tag VBG is rarely picked here by the automatic tagging programs. But both nouns and adjectives occur attributively. Compare:
Distinguishing between NN and JJ in attributive position is often problematic (see 7.9), and this is especially true of -ing forms. It is indeed debatable whether we can, or should, draw a borderline between NN and JJ with attributive -ing forms. But since such a distinction is traditionally made in grammars and dictionaries and since there are clear cases of both categories, an attempt has been made to distinguish them. Criteria:
1. If the -ing form is clearly established in other positions as NN rather than JJ, or vice versa, these tags can safely be extended to attributive position. Compare:
2. A preceding modifier may point in the direction of NN or JJ. Compare:
3. Coordination with clear nouns or adjectives is helpful in many cases; cf. (96)-(97) above. Similarly, substitution by clear nouns or adjectives can often be a good guide. Compare:
4. The principal criterion has been paraphrase. A JJ form can be paraphrased with a following relative clause: 'who/which BE... JP or 'who/which VB...'. NN can be paraphrased as: 'for/of ...-ing'. Compare:
NN |
JJ |
|
finishing school A37:35 |
finishing touches E04:51 |
|
living arrangements F02:48 |
living animals G02:137 |
|
living space B26:112 |
the living brain D14:77 |
|
retiring age H27:57 |
the retiring Mayor A30:15 |
|
a sitting position P04:90 |
sitting targets A23:98 |
|
a standing place in the |
his humanistic standing |
|
spiritual world G32:119 |
figures C07:109 |
|
standing room only P01:86 |
standing committee A11:183 |
|
travelling dress G39:3 |
a travelling zoo G54:84 |
|
working hours A30:115 |
working mother A35:196 |
|
working space E27:94 |
the working population B15:61 |
Some forms allow both paraphrases, e.g.:
The tendency in such cases has been to opt for JJ, but there is no doubt some inconsistency on this point.
Complex attributive -ing forms have been treated in the same way as simple ones. Compare:
Where there is no hyphen, each word must be tagged individually, as in:
The tag VBG is not used in cases like (109), as the 'object' precedes the -ing form and the sequence is therefore quite different from that of a clause with a verb followed by its object.
There are many problems with our paraphrase criteria. Some forms which we would like to tag
JJ do not fulfil the paraphrase criterion for JJ, e.g.:
In (110) it is definitely not the newspaper which disapproves. It is less certain whether the paraphrase criterion fails with (111), since loving does occur as an adjective in predicative position, as in: Her look was loving. In such cases an attribute is transferred from a person to a thing associated with the person, often a part of the body. Since such 'transferred' modifiers are also found with bona-fide adjectives (eager eyes), the tagging JJ is the natural one.
Many forms have been tagged NN, although they do not satisfy our paraphrase criterion for NN (provided that the adjective criterion also fails). Examples:
NN-tagging is questionable in examples like (112), since sporting does not occur in clearly nominal positions. The form is also listed as an adjective in many dictionaries (though it lacks clear adjective characteristics).
English -ing forms are versatile in the functions they can fulfil and the meanings they can express. In many cases it is difficult to draw a borderline and, more often than elsewhere, we had to sacrifice consistency of classification and fall back upon our 'follow-the-tagger' principle (Section 6).
