Readme

August 5, 2003
 Yusuke Kubota, Kyoko Takano, Camille Yamada & Yuka Tateisi

Contents

1. Introduction
1. Outline of Fixing Procedure
2. Segmentation of Sentences
1. Indicators of  Fixed  Segments
2. Periods
3. Parentheses and Commas
4.  Quotations and Apostrophes
5. Hypens and Minuses
6. CANNOT
7. Ratio
8. Slashes
9. Bibliographic Information
3. Tagging POS
1. Symbols for Aiding Manual POS-fixing
   
1. "!" assigned by the tagger
2. Automatically inserted alternatives POSes and comments
1. Alternative POSes (word/POS1.pos2)
2. Automatically inserted comments (word/POS[comment])
3. Asterisks for marking technical terms
2. Indicatiors of Fixed Tags
3. JJ or NN
1. Words to be tagged as JJ
2. Words to be tagged as JJ or NN depending on the context (expect -ing  words)
3. Words to be tagged as NN
4. JJ or VBN
5. JJ, NN or VBG
1. "Binding"
2. -ing  words other than "binding"
6. Symbols
1. Symbols to be tagged as NN(S)
2. Symbols not to be tagged as NN(S)
7. NN or NNP
8. Idioms
9. Latin Words etc
10. Other Foreign Words
11. Suffixes
12. Acronyms
13. NN(P) or NN(P)S
1. Acronyms and abbreviations of units
2. Other cases
14. Slashes
    
15. Mistakes in the original text
16. Misc

• Notes
• -ed
• -ing
• PREPOS data
  

1 Introduction

1.1 Outline of the fixing procedure

Words should be tagged according to the Penn Treebank Manual (henceforth PTB Manual) unless stated otherwise in this document. Annotators are given files in which words are automatically tagged by a POS-tagging program. You should fix (i) segmentation (tokenization) of words and sentences and (ii) POS tags assigned to words at the same time.

2 Segmentation of sentences and words

POS-tagged files should contain one sentence per line. Words in a sentence should be separated by spaces. Special attention should be paid to periods, parentheses, commas and quotations.

Any inconsistent segmentations of a given word in the same abstract should be left as they are unless the inconsistency occurs around hyphens, parenthesis and commas. Inconsistencies found around hyphens, parenthesis and commas should be fixed according to the principles stated in sections 2.3 and 2.5.

E.g:

HpaII and Hha I

"HpaII" and "Hha I" are inconsistently segmented here (one with a space and the other without one) but should be left as they are since there are no hyphens, parenthesis and commas concerned.

2.1 Indicators of fixed segmentations

When you have fixed the segmentation of a given word or sentence, put one of the following marks to indicate the fixed place for later reference.

The "cut" mark should be inserted between the separated two words. The "concatenated" mark should be inserted immediately after the concatenated word.

Marks indicating places where something has been fixed such as "#" and "=" or those indicating uncertainty on the part of the annotator such as "@@" and "?" are introduced for the purpose of facilitating error checking at the final stage of POS-fixing. So, there is really no precise guideline for where to put them and where not to put them or the order among them when more than one marks are assigned to one word. Annotators should not be troubled with these questions and should make decisions according to their own judgements.

Some examples follow:

To cut one word into two:

abcde/NN → ab/NN # cde/NN
abcdefgh/NN → ab/JJ # cd/JJ # ef/JJ # gh/NN

To concatenate two words into one:

ab/JJ cd/NN → abcd/NN =
ab/NN cd/NN ef/NN gh/NN → abcdefgh/NN =

To fill in a dropped word:

B/NN! → B/Rel/NN! *

To cut one sentence into two:

... by PCR./NN We have investigated ...
↓
... by PCR/NN # ./.! #
We have investigated ...

To concatenate two sentences into one:

... 0.23/CD vs/NNS ./.
3.7/CD +/-/CC ...
↓
... 0.23/CD vs./CC! = 3.7/CD +/-/CC ...

2.2 Periods

Sentence-final periods should be separated from the preceding words.

E.g.:

Nakazawa/NNP met/VBD Yatabe./NNP → Nakazawa/NNP met/VBD Yatabe/NNP # ./.

Other periods should be concatenated with the preceding words.

E.g.:

Dr/NNP ./. Charcot/NNP → Dr./NNP = Charcot/NNP

Note that when an acronym ending in a period (e.g. L.I.S.) occurs at the sentence-final position, the last period should also be separated from the word.

E.g.:

They/PRP are/VBP members/NNS of/I N L.I.S./NNP → ... L.I.S/NN P # ./.

Segmentations of initials, even if they are inconsistent, should be left as they are. (E.g., "J.M . Folon" should not be fixed as "J.M.Folon"or"J . M . Folon". Likewise, "J . M . Folon" should not be fixed as "J.M . Folon" or whatever.) However, in cases like "R.and" where it is clear that the initial is mistakenly concatenated with the next word, the two words should be separated as in "R . and".

Note that periods are used for various purposes other than marking sentence boundaries, such as decimal points, part of EC numbers (e.g. EC.1.2.3.4) and part of substance names (e.g. PU.1). If you are uncertain whether or not to cut a sentence at a given period, refer to the original .txt file and keep the segmentation consistent with it.

When a sentence is incorrectly separated into two in the original .tx t file, the two sentences should be concatenated into one but the word with a period (e.g. E . in the following example) should not be concatenated with the next word in the corrected POS-tagged file.

E.g.:

E/FW ! ./. histolytica/FW !
↓
E./FW ! # = histolytica/FW ! (not E.histolytica/FW)

2.3 Parentheses and commas

Parentheses and commas appearing inside substance names should be concatenated with preceding and following words.

E.g.:

• 2,3,7,8-tetrachlorodibenzo-p-dioxin/NN
• 1/CD alpha,25(OH)2D3/NN
• [3H]-Dexamethasone/NN
• gamma(c)/NN

Parentheses and commas appearing inside numerical expressions or formulas should also be concatenated with preceding and following words.

E.g.:

• 10(6)/CD ("10⁶")
• 10(-10)-(-6)/CD (range; "from 10¹⁰ to 10⁶")
• t(15;17)/NN (translocation)
• (q11;q32)/NN (location in gene)
• inv(14)(q11;q32)/NN (inversion at location in gene)
• CD4(+)/JJ (CD4-positive)

Other parentheses and commas should be separated from preceding and following words. Note in particular that parentheses enclosing abbreviations of substance names should be separated from preceding and following words.

E.g.:

Interleukin/NN (/( IL/NN )/) -6/CD

Tokens like "10(-4)/bp" should be regarded as a single expression designating a unit and hence should be segmented as a single word and tagged as NN. ("10(-4)" means 10⁴)

2.4 Quotations and apostrophes

The rules in this section, except the first one, are tacitly observed in the Penn Treebank corpus, although they are not explicitly stipulated in the PTB manual.

• "'" in "5'" or"3'" (terminus of gene) should not be separated from the numbers.
• "'s" and "'" used as possessive suffixes should be separated from the preceding word (e.g. "Hodgkin/N N 's/POS").
• "'" indicating the beginning of a citation should be replaced by "`" and should be separated from the next word.
• "'" indicating the end of a citation should be separated from the preceding word.
• """ indicating the beginning of a citation should be replaced by "``" (two backquotes) and should be separated from the next word.
• """ indicating the end of a citation should be replaced by "''" (two single quotes) and should be separated from the preceding word.
• The order between the sentence final period and "'" or """ indicating the end of a citation should be left as it is.

2.5 Hyphens and minuses

Hyphens used as dashes should be separated from preceding and following words.

E.g.:

This-also profound-blah blah blah → This # - # also profound # - # blah blah blah .

Words separated at hyphens inside them (e.g. "Hughling-Jackson") should be concatenated unless the fragment with the hyphen constitutes a conjunct (e.g. "up-"in"up- and downregulation"). Some examples of conjunctions follow:

• (both ) ... and ...
• (either ) ... or ...
• (neither ) ... nor ...
• (from ) ... to ...
• plus ...
• vs ...

Note that there are words ending with a minus. Examples of words ending with a minus are:

• names of minus ions (should be tagged as NN)
  • OH-/NN
  • Cl-/NN
• +/-(should be tagged as CC)
• expressions indicating that something is lacking (should be tagged as JJ)
  • CDnn-(where "nn" stands for any given number)
  • FCS-
  • IgA-

"-" used in expressions indication ranges or locations in genes or used as parts of substance names should all be concatenated with preceding and following words.

• examples of ranges:

  8- 15 min → 8-15/CD min/NN

• examples of positions and locations:

  6p26- 28 → 6p26-28/NN

• examples of substance names containing "-" inside them:

  IL- 6 → IL-6/NN

Expressions of the form "Substance Name (Abbreviation)-Number" should be segmented and tagged as "Substance Name/NN (/( Abbreviation/NN )/)# -Number/CD".

E.g.:

interleukin (IL)- 6 → interleukin/NN (/ ( IL/NN )/) # -6/CD

Similar example:

Nuclear Factor (NF)-kapp a B → Nuclear/JJ Factor/NN (/ ( NF/NN )/) # -kappa/NN B/NN

When "-" follows ")" in expressions of the form "Substance1 (XXX)-and Substance2 (YYY) -XXXed", (where "XXX" and "YYY" supply some information related to Substance1 and Substance2, respectively) the hyphen between ")" and "and" should be separated from preceding and following words and should be tagged as ":".

E.g.:

lipopolysaccharide/NN (/( LPS/NN )/) -/: and/CC N-formyl-methionyl-leucyl-phenylalanine/NN (/( FMLP/NN )/) -stimulated/JJ

A space appearing inside a part of a substance name enclosed by parentheses should be converted to an under bar ("_"). For example, "1-(5-Isoquinoliny l sulfonyl)-2-methylpiperazine" should be converted to "1-(5-Isoquinolinyl_sulfonyl)-2-methylpiperazine". Spaces outside parentheses should be left as they are unless they are immediately preceded or followed by a hyphen.

2.6 CANNOT

"cannot" should be split into "can" and "not" in accordance with the PTB Manual and should be tagged as "can/MD not/RB".

2.7 Ratios

"Substance Name:Substance Name" should be segmented as a single word and tagged as NN.

E.g.:

colombus/NN :/: milk/NN =/JJ 1:3/CD → colombus:milk/NN = =/JJ 1:3/CD

E.g.:

1/CD :/ : 3/CD → 1:3/CD =

2.8 Slashes

Parentheses and slashes in expressions of the form "Substance1 (XXX)/Substance2 (YYY)"(where "XXX" and "YYY" supply some information related to Substance1 and Substance2, respectively) should all be separated from preceding and following words. In this case, "/" should be tagged as ":".

E.g.:

cAMP response element binding protein ( CREB ) //: activating transcription factor- 1 ( ATF- 1 )

2.9 Bibliographic information

Parentheses and semicolons appearing as part of bibliographic information that are concatenated with preceding and following words in the original .txt file should be separated from preceding and following words.

E.g.:

Immunity 1999 Mar;10(3):39 9 → ... Mar/NNP ;/: 10/CD (/ ( 3/CD )/) :/: 399/CD

However, expressions indicating ranges of pages should be treated as a single unit and should be tagged as CD (e.g. "399-420/CD"). This is in accordance with the rule that ranges should be segmented as a single word (section 2.5). Pointers to references like "[1]" and "(1)" appearing at the end of a sentence should be segmented and tagged like "[/ ( 1/C D ]/)".

3 Tagging POS

Technical terms, Latin words and idiomatic expressions that frequently appear are automatically assigned correct tags according to the Term List. Hence, if you are uncertain whether to fix the tags of such words, you should rather leave them as they are. When you are uncertain about the tags of ordinary words, try to determine them by the following procedure:

1. Follow the guidelines in this document. If it doesn't help,
2. follow the guidelines in the PTB Manual. If it doesn't help,
3. consult the Longman Dictionary of Contemporary English (LDOCE), 3rd edition. If it doesn't help,
4. consult Kenkyusha's English-Japanese Dictionary for the General Reader, 2nd edition. If you are still uncertain,
5. refer to previous corpus. Bear in mind that you can refer to previous corpus only for the purpose of confirming some rule or agreement that you have become unsure of. If you want to refer to previous corpus for other purposes, consult other annotators and persons concerned beforehand. If you are still uncertain,
6. send an email to other annotators and persons concerned. The WSJ Corpus might be consulted at this stage if considered appropriate.
7. Discuss with other annotators and persons concerned to establish an agreement.

3.1 Symbols for aiding manual POS-fixing

Various symbols are inserted to the POS-tagged file for the purpose of aiding the manual POS- fixing procedure. The symbols can be classified into two broad categories: (i) those marking words whose POS tags don't have to be checked and (ii) those marking words whose tags should be determined by specific principles.

This section describes in detail what these symbols mean and what they do not mean. (Note that correct POS-fixing is impossible when you are unsure how to read such symbols. Since some of the symbols are inevitably quite confusing and misleading, even an experienced annotator should pay close attention not to misunderstand the meanings of these symbols.)

3.1.1 "!" assigned by the tagger

POS tags already followed by "!" (e.g. "NF-kappaB/NN!", "interleukin-1/NN!") are assigned by the POS tagger according to the Term List. Hence, you don't usually have to fix such tags. However, if you find any apparent errors with these tags, correct them and put the mark "!!" (if you are confident about your correction) or "!?" (if you are not confident). If the error in question is a simple careless mistake that can be easily fixed, report it immediately by email. Also, when you find words that should be registered in the PREPOS list but is not yet registered, report it by email.  Special attention should be paid to the following cases:

1. past participles tagged uniquely as VBN by the POS tagger¹
   A past participle tagged uniquely as VBN by the tagger should nevertheless be tagged as JJ when it appears between nominal expressions, and should be tagged as VBD when it is used as a past main verb.

   E.g.:

   We/PRP investigated/VBD how/WRB these/DT stimuli/NNS affect/VBP mitogen/NN activated/JJ!! protein/NN (/( MAP/NN )/) kinases/NNS ./.

   Thus/RB ,/, we/PRP demonstrated/VBD that/IN IL-5/NN activated/VBD!! the/DT Jak/NN 2-STAT/NN 1/CD signaling/NN pathway/NN in/IN eosinophils/NNS ./.

   the POS tags of `s'
   `s' is tagged as NNS by the POS tagger (since it should be so tagged in vast majority of cases). However, it should be tagged as VBZ when it is used as a third person singular present suffix, and as NN when it abbreviates for "second" (unit).

3.1.2 Automatically inserted alternatives

POSes and comments Letters followed by a tilde `~' or a comment [...], inserted immediately after the POS tag by the script tilde.pl, calls attention to correct POS tags. Comments and signs inserted by tilde.p l are all removed by the script rmtilde.p l after the manual POS-tagging process is completed.

3.1.2.1 Alternative POSes (word/POS1.pos2)

Some words should be assigned different POS tags in accordance with the contexts in which they appear. In such cases, a candidate alternative POS pos2 is sometimes inserted in the lower case preceded by a tilde after the original POS POS1. In other words, word/POS1.pos2 means that although word was originally tagged as POS1 by the tagger, it might as well be pos2. The correct POS tag should be determined by the annotator according to the context.

E.g.:

upstream/RB~jj This means that the POS of `upstream', which was originally tagged as RB, might be JJ.

3.1.2.2 Automatically inserted comments (word/POS[comment])

When a comment [comment] is inserted immediately after the POS tag, fix the POS according to it.

• comments for -ed words .
  Criteria that are most likely to be applied judging from the words and context (described in 3.3) are given in [ ]. Note that not all criteria are given. In some case, a criterion in [ ] should not be adopted because of the existence of more strong one which is not inserted as a comment (e.g. "when a given past participle modifies a nominal expression and is preceded by a nominal expression modifying it or is hyphenated, it should be tagged as JJ").

  [VBN before by+agent]

  When a given past participle is a complement of the be-verb or modifies a preceding nominal expression and is followed by a by-phrase designating the agent, it should be tagged as VBN. When applying this rule, make sure that the noun after `by' really refers to an agent.

  [-ed=jj]

  If a past participle is followed by "state", "form", "pattern" or "fashion", this comment is inserted to direct the annotator's attention to the rule that past participles are tagged as JJ in such environments.

  [JJ if opposite of XX]

  Expressions of the form "non", "un" or "in" with the meaning "opposite of XXX" or "not XXX" should be tagged as JJ. You should confirm that "non", "un", and "in" are actually used in the meanings "opposite of XXX" or "not XXX" since this comment is automatically inserted if the word begins with either of these strings (without regard to the meanings of such words).

  [JJ if meaning=before]

  Expressions of the form "preXXX" (in which the prefix "pre" carries the meaning "before") should be tagged as JJ.

  [POS|POS]

  This pattern is the reminder of the ED list.

  1. [vbn|vbn] in both the prenominal position and other environments, should be tagged as VBN
  2. [jjv|bn] in the prenominal position, should be tagged as JJ; in other environments, should be tagged as VBN
  3. [jj|jj] in both the prenominal positions and other environments, should be tagged as JJ
  4. [see ed-list] The POS of -ed words with this comment is determined only in limited environments (e.g. `JJ if prenominal'). Follow the ED list if there is a description. In the future, such words should be included in one of the three categories given above.

• comments for -ing words
  There are three kinds of comments about the -ing form: [-ing=nn], [-ing=jj] and [-ing=vbg]. Every comment can be classified into the following two categories, where clause 1 has priority over clause 2. Note that some criteria in 3.4 (e.g. `when "NN XXing" is a prenominal modifier or behaves as a nominal expression, "XXing" should be tagged as NN') should be given priority over these comments.
  1. entries in the INGlist

     E.g.:

     signaling/NN[-ing=nn] pathways/NNS

  2. POS tags determined by the following nouns

     E.g.:

     binding/NN[-ing=nn][in principle] activity/NN

     The comment [in principle] indicates the possibility that the -ing word and the following word are not in apposition.
• Other comments

  [NNS?]

  The POS of words with this comment might be NNS.

  [from postpos]

  This comment indicates that the POS tag assigned by the tagger or the PREPOS script was corrected by POSTPOS.

  [if prenominal:JJ]

  Words with this comment should be tagged as JJ in the prenominal position.

  [even if prenominal:NN]

  Words with this comment should be tagged as NN even in the prenominal position.

  [even if noun:JJ]

  Words with this comment should be tagged as JJ even if it is used as a noun.

  [even if adverb:JJ]

  Words with this comment should be tagged as JJ even if it is used as an adverb.

  [if conj:VBN]

  Words with this comment should be tagged as VBN if they are used like a conjunction.

  [if conj:VBG]

  Words with this comment should be tagged as VBG if they are used like a conjunction.

  [if obviously plural:NNS]

  Words that end in -s in both singular and plural forms should be tagged as NN by default. However, they should be tagged as NNS if they appear at positions where only plural forms can appear. (Refer to section 3.13.) E.g.: by all means/NNS[if_obviously_plural:NNS]

  [NN|VBX]

  Words with this comment can be either a noun (NN) or a verb (VB or VBP) depending on the context. E.g.: focus/NN[NN|VBX]

3.1.3 Asterisks for marking technical terms

Some words are surrounded by concatenations of two asterisks (e.g. **TNFRI**/NN). As for such words, you don't have to care about whether the POS is correct or not.

Asterisks are attached to words which do not appear in ordinary dictionaries. As a result, many of them are technical terms. However, asterisks are eliminated from words which are composed of combinations of more than one elements of 1-3 (e.g. E1A, PCOS, LTalpha), because they might be NNS. Instead, the comment [NNS?] is attached to such words.²

1. a capital letter
2. an English equivalent of a Greek alphabet
3. a numeral

3.2 Indicators of fixed tags

Each POS tag should be followed by one of the following marks:

Note that

• when you correct a POS tag already followed by !, follow the procedure stated in section 3.1.1. 
• when you fix the segmentation and POS tag of a given word or phrase at the same time, put ! only when the POS tag of the last word changes.

  E.g.:

  Alpha/NN CD3/NNS → AlphaCD3/NN! =

  Alpha/NN CD3/NN → AlphaCD3/NN =

• ungrammatical sentences should be preceded by the mark "@@". If the ungrammaticality can be attributed to a specific word (e.g. in the case of agreement mismatch), it might be helpful to put "@@" also to the word in question immediately after its POS tag.
• if you feel uncertain whether or not to put "@@" or"?", put them rather than omit them.
• as noted in section 2.1, there is no precise guideline for the placement of marks indicating uncertainty on the part of the annotator. You should make decisions according to your own judgements if there is no explicit rule in this document that is applicable to the specific case you are dealing with. 

3.3 JJ or NN

3.3.1 Words to be tagged as JJ

• Hyphenated words in the prenominal position should generally be tagged as JJ except those which are already known to be a substance name (and the like) and hence should be tagged as NN. Annotators need not consult other annotators or persons concerned about such cases. Hyphenated substance names will be picked up and converted to NN at the final stage of refixing. Hyphenated words that appear as nominal heads should be tagged as NN. 

  E.g.:

  • granulocyte-macrophage/JJ lineage/NN
  • protein-protein/JJ interactions/NNS
  • gel-shift/JJ
  • pro-B/JJ
  • mobility-shift/JJ

• Expressions in which two words whose independent POSes are JJ or NN are concatenated by a slash and which function as prenominal modifiers should be tagged as JJ. Refer to section 3.14. (In such cases, the function of slashes are almost identical to that of hyphens.) 
• "control" used in the meaning "of the control group"
• "human" modifying a nominal expression
• "adult" modifying a nominal expression
• "net" in "net weight" or "net effect"
• "immune" in "immune system"
• all-trans, all-cis
• unilineage, multilineag e
• "n-fold" (where "n" stands for any given number) should be tagged as JJ if it modifies a nominal expression or appears as a complement of the be-verb. Those used adverbially should be tagged as RB.
• Expressions of the form "XXX. " (or "XXX+") with the meaning "XXX negative" (or "XXX positive") should be tagged as JJ. "CD+" and "CDnn!" (where "nn" stands for any given number, e.g. CD4+) immediately followed by cell names such as "T cell" fall under this category.
• Expressions of the form "Substance Name-/-", "Substance Name-/+", "Substance Name+/-" and"Substance Name+/+" modifying cell names (e.g. "Rel-/- T cells", "Rel+/+ T cells") should be tagged as JJ. If such an expression appears as two words like "Rel -/-", the two parts should be concatenated into one.

  E.g.:

  kappa/NN B/NN alpha-/-/JJ

• Expressions of the form "5'" (or "5'-") and "3'" (or "3'-") should be tagged as JJ. If they function adverbially, they should be tagged as RB.
• Expressions of the form "nn-unit" (where "nn" stands for any given number, e.g. nn-kD a or nn-kilodalton) modifying nominal expressions should be tagged as JJ.
• helix-loop-helix/JJ . 5'-flanking/J J
• 5'/JJ flanking/JJ
• "mean" used in the meaning "average"
• "few", "little", "former" and "latter" should be tagged as JJ even when they appear as nominal heads.
• "male" and "female" in prenominal positions
• "IgH" in prenominal positions ("IgH" stands for "immunoglobulin heavy-chain" or "immunoglobulin heavy chain". It only appears in prenominal positions.)
• non-Hodgkin/JJ 's/POS (disease)
• NN upstream/JJ of . . . (This expression can be paraphrased as "NN which is upstream of . . . ". Hence, it is regarded as a reduced relative and should be tagged as JJ.)
• {situated,located}/JJ upstream/JJ of ...
• terminal/JJ
• tandem/JJ .
• "split" in the prenominal position
• prior/JJ to . . . (even if used adverbially)

3.3.2 Words to be tagged as JJ or NN depending on the context (except -ing words)

• anti-XXX (where XXX stands for a given word)
  • should be tagged as JJ when it directly modifies a nominal expression.
  • should be tagged as NN when it appears in an environment where only nominal expressions can appear (e.g. as an object of a preposition). (In this environment, this expression is used to refer to "an antibody to XXX".)
  • Expressions of the form "anti-XXX {activated,inhibited}" modifying a nominal expression should be tagged as "anti-XXX/NN {activated,inhibited}/JJ"
  • "Ehrlichia chafeensis" is a scientific name of bacteria and is tagged as FW. . Hence, "Anti-Ehlichia" and "Anti-E" should be tagged as FW|JJ. 
    
• Expressions of the form "XXX-ant" (e.g. immunosupressant, recombinant, mutant, variant) whose meaning is either "has an ability to XXX" or "something that XXXes" should be tagged as
  • JJ when it directly modifies a nominal expression.
  • NN when it appears in an environment where only nominal expressions can appear . (e.g. as an object of a preposition).
  • Expressions of the form "XXX-ant {activated,inhibited}" modifying a nominal expression should be tagged as "XXX-ant/NN {activated,inhibited}/JJ".
• The following words should be tagged as NN when they function as nominal heads (including the cases of apposition) and as JJ when they function as prenominal modifiers: (This list should be expanded to include the list of suffixes; can "-oid" be generalized?)
  • autocrine
  • endocrine
  • male
  • female
  • extra
  • promoter-enhancer
  • diploid
  • tetrapoid
  • transXXX
  • preXXX
  • postXXX
  • proXXX

3.3.3 Words to be tagged as NN

• substance names
  Prenominal unknown words that match none of the following patterns can be tentatively regarded as substance names and should be tagged as NN. If the unknown word in question is not circumfixed by asterisks, they should be assigned pending marks ("^"). Whether they are really substance names will be checked at the final stage of re-fixing.
  • hyphenated words .
  • words containing slashes
  • words ending in + or
  • words with the prefixes "anti-", "pre-" and "pro-"
  When a hyphenated word as a whole functions as a substance name (and the like), it should be tagged as NN (e.g. T-cell/NN, AP-1/NN).
• cell names that modify nominal expressions (e.g. "IgM" in "Ig M receptor")
• chemical elements and ions (e.g. Ca2+, Cl-)
• abbreviations . Note that Roman numerals should be tagged as CD.
• "surface" used in the meaning "on the surface of something" modifying a nominal expression (So determined because LDOCE has no adjective entry for this word.)
• "patient" in the meaning "person receiving medical treatment" modifying a nominal expression.
• For the distinction between nouns and present participles or gerunds of verbs, refer to section 3.5. . animal/NN model ("animal" should be tagged as JJ in other prenominal positions.) . Time expressions should be tagged as NN even if they function adverbially. (From PTB Manual)

  E.g.:

  little is known about the regulation of expression of genes expressed ` late/JJ ' (three or more days) post-stimulation/NN

3.4 JJ or VBN

1. When a given past participle modifies a nominal expression and is preceded by a nominal expression modifying it or is hyphenated, it should be tagged as JJ. (From PTB Manual)

   E.g.:

   • 10/CD genetic/JJ probe-based/JJ tests/NNS
   • a/DT government-approved/JJ safety/NN seat/NN
   • airline/NN leveraged/JJ buy-outs/NNS
   • Thomson/NN consolidated/JJ earnings/NNS

2. When a given past participle is a complement of the be-verb or modifies a preceding nominal expression and is followed by a by-phrase, it should be tagged as VBN. (From PTB Manual)
3. If the past participle is followed by "state", "form", "pattern" or "fashion", it should be tagged as JJ.
4. Expressions of the form "nonXXX", "unXXX" or "inXXX" with the meaning "opposite of XXX" or "not XXX" should be tagged as JJ.
5. Expressions of the form "preXXX" (in which the prefix "pre" carries the meaning "before") should be tagged as JJ.
6. Refer to the PREPOS list.
7. Refer to the ED list that is appended to this document. The POS of a word that doesn't appear in the ED list should be determined by the following criteria. When the POS is determined, the word will be added to the ED list if it is considered to be appropriate. (As usual, among the following criteria, priority should be given to higher items over lower ones.)
   1. Words that should be assigned different tags for different meanings:
      • disturbed, depressed: when used to refer to a certain human mental condition, should be tagged as JJ; otherwise, should be tagged as VBN.
      • cripped, inflamed: when used to refer to a condition of (the body of) a living thing, should be tagged as JJ; otherwise, should be tagged as VBN.
   2. -ed words whose base forms are unambiguously intransitive should be tagged as JJ.
   3. -ed words that has no corresponding base verb form (e.g. threaded) should be tagged as JJ.
   4. If the past participle is an expression that refers to a position or place (e.g. located, positioned), it should be tagged as JJ.
   5. If the past participle refers to a result of a chemical reaction (e.g. phosphorylated, ionized; activated, inhibited), it should be tagged as VBN.
   6. -ed words with adjective entries in LDOCE should be tagged as JJ as long as the meanings are not significantly different from the dictionary definitions.
   7. If the past participle refers to an event it should be tagged as VBN; if it refers to a state, it should be tagged as JJ. (From PTB Manual)
   8. -ed words with prefixes should generally be tagged in the same way as those without prefixes. (E.g.: coactivated and colocated should be tagged in the same way as activated and located, respectively; the former should generally be tagged as VBN and the latter should in general be tagged as JJ. Note, however, that there are some exceptions to this generalization, e.g., related/J J vs. correlated/VBN.)
8. If "XXXed NN" can be paraphrased as "NN that is (are) XXXed", "XXXed" should be tagged as VBN. If not, it should be tagged as JJ.
9. 9. If you feel uncertain or if you want to refer to the WSJ Corpus, report the case by email. Decisions will be made by discussion among annotators and persons concerned.

3.5 JJ, NN or VBG

3.5.1 "Binding"

The POS of "binding" should be determined by the following procedure. (Priority should be given to higher rules over lower ones.) If you are uncertain about your decision, put the mark "^" immediately after the POS tag as in "NN^". (If you cannot determine the POS of a given instance of "binding" uniquely, you should specify the (logically and contextually) possible alternatives by the vertical bar notation as in "NN|VBG" and put "^" after the POS tag. If you cannot even specify tags like "NN|VBG", just put "^" on the word. If you can determine the tag uniquely with confidence, "^" is not necessary.)

1. Follow the ING list and the PREPOS list.
2. When "binding" is hyphenated, apply the guidelines about hyphenated words in this document.

   E.g.:

   To gain a better understanding of the mechanism of binding to the human mineralocorticoid receptor (hMR), we developed a new monoclonal antibody (mAb) raised against the hormone-binding/JJ domain (HBD).

3. When "binding" is in the head-of-subject position without any preceding determiner, possessive pronoun ("its"), adjective or adverb, it should be tagged as NN.

   E.g.:

   this region is also involved in NFAT binding to activated calcineurin, since binding/NN is inhibited by an NFAT peptide spanning the calcineurin docking site on NFAT.

4. When "binding" has a dependent determiner or adjective as a preceding word, it should be tagged as NN.

   E.g.:

   In gel retardation analysis, FK506 had little effect on the binding/NN to the AP-1 site of PMA/ionomycin-induced nuclear factors, which were recognized with anti-JunD or c-Fos antibody.

5. When "binding" is immediately followed by an of -phrase, it should be tagged as NN.

   E.g.:

   When we analyzed the nature of STAT proteins capable of binding to IL-2Ralpha, pim-1, and IRF-1 GAS elements after cytokine stimulation, we observed binding/NN of STAT1, STAT3, and STAT4, but not STAT5 to all of these elements.

6. When "binding" functions as an object of an immediately preceding preposition, it should be tagged as NNVBG. However, if "binding" appears in the expression "capable of binding", it should be tagged as VBG.

   E.g.:

   We propose that HIV-1 tat may inhibit MIP-1 alpha expression by inducing MNP-1 expression in T-cells, probably by either competing with MNP-2 for binding/NN|VBG^ to the MIP-1 alpha promoter or by sequestering it into inactive forms.

7. When "binding" is immediately preceded by an NN(S),
   • if it is immediately followed by a to-phrase, it should be tagged as NNjVBG.

     E.g.:

     Characterization of a new tissue-specific transcription factor binding/NN|VBG^ to the simian virus 40 enhancer TC-II (NF-kappa B) element.

   • if it is immediately followed by an NN, apply the rule 4 in section 3.5.2. If you cannot determine the tag by this procedure, put "^" on "binding". Be careful to check out whether "binding" functions as an object of a verb or not. In the case when it is an object of a verb, it should be tagged as NN|VBG.
   • if it is not followed by anything (i.e. if "NN binding" forms a phrase by itself), "binding" should be tagged as NN.

3.5.2 -ing words other than "binding"

1. When a given present participle modifies a nominal expression and is hyphenated, it should be tagged as JJ. (From PTB Manual)⁴ Note that when the present participle in question is used in apposition to the following noun as in "DNA-binding/NN activity"⁵, it should not be regarded as a prenominal modifier, and hence, this rule should not be applied. 
   
2. When "NN XXing" is a prenominal modifier or behaves as a nominal expression, "XXing" should be tagged as NN. (From PTB Manual)
3. Refer to the PREPOS list.
4. In some cases, the POS is determined by the following noun. . -ing/NN {activity, affinity, capacity, characteristics, effect, event, property, function, mechanism}
   When the -ing word is in apposition to the following noun, it should be tagged as NN. -ing words are often used appositively when followed by one of the nouns listed above. However, we have found one instance of "-ing event" in which the -ing word is not in apposition to the following noun (i.e. event). Hence, be careful to consider the context when applying this rule. . -ing/JJ {form, pattern, fashion, state} (the same as "-ed/JJ form", etc.)
5. Refer to the ING list that is attached to this document.
6. -ing words with adjective entries in LDOCE should be tagged as JJ as long as the meanings are not significantly different from the dictionary definitions.
7. Expressions of the form "preXXXing" in the prenominal position should be tagged as JJ. It is not likely to appear as a complement of the be-verb.
8. If "NN1 XXing NN2" can be paraphrased as "NN1 that XXs NN2", "XXXing" should be tagged as VBG. (From PTB Manual)
9. If "XXing NN" can be paraphrased as "NN that XX(es) (something)", "XXXing" should be tagged as VBG. (From PTB Manual)
10. If you encounter words that you have not seen before and feel uncertain about its POS, report the case by email. Decisions will be made by discussion among annotators and persons concerned.

3.6 Symbols

3.6.1 Symbols to be tagged as NN(S)

3.6.2 Symbols not to be tagged as NN(S)

• Roman numerals (I⁶, II, III, ...): CD
• List numbers⁷: LS
• Field indicators (UI, TI, and AB): LS
• "+", "-" and "+/-" used as conjunctions: CC
• --> (right arrow): TO
• = (equal sign): The tag of the most appropriate paraphrase (e.g. "equal/JJ to", "is/VBZ equal to", "was/VBD equal to", "being/VBG equal to") should be chosen. If you are uncertain, simply tag it as JJ.
• x (multiplication sign): CC
• <, > (less-than sign and greater-than sign): The tag of the most appropriate paraphrase (e.g. "is/VBZ less than", "was/VBD less than", being/VBG less than") should be chosen.

  E.g.:

  Patients were classified as steroid sensitive (SS) if their. morning FEV1 increased >/RBR 30%. However, it was still lower than that in healthy subjects (P < 0.01).

3.7 NN or NNP

• names of authors of abstracts
• titles of journals, proceedings, and books. Plural forms should be tagged as NNPS (e.g. "J./NNP Immunol./NNP Methods/NNPS").
• initials of names of patients
• country names
• names of organizations (hospitals and laboratories)
• names of months (January, February, . . . )

3.8 Idioms

3.9 Latin words, etc.

• Components of scientific names of living organisms should be tagged as FW (e.g. "Saccharomyces/FW cerevisiae/FW").
• English equivalences of Greek letters should be tagged as NN (e.g. alpha/NN).
• Components of idiomatic expressions originating from foreign languages that are frequently used in the literature of biology should be tagged as FW.

  E.g.:

  • in/FW vitro/FW
  • in/FW vivo/FW
  • de/FW novo/FW
  • in/FW situ/FW

  Other idioms of foreign origin are either tagged as FW or assigned tags that correspond to their functions according to the WSJ Corpus.
  • per/FW se/FW
  • et/FW al./FW
  • etc./FW i.e./FW (or, "i./FW e./FW")
  • e.g./FW (or, "e./FW g./FW")
  • ca./FW
  • vis-a-vis/FW
  • bona/FW fide/FW
  • via/IN
  • versus/CC
  • vs./CC
  • per/IN (except "per/FW se/FW")
  • plus
    • NN when used as a noun as in "a plus"
    • CC when used as a conjunction as in "A plus B"
  • vice/RB versa/RB .
  Do not try to keep the segmentations of words with periods consistent but rather leave them as they are. (E.g.: "i. e." or"e. g." should not be fixed as "i.e." or "e.g.".) . When an English word is hyphenated with a foreign word, the whole concatenated word should be tagged as FW|XX where XX stands for the tag of the English word if the word it is hyphenated with were not a foreign word (e.g. in/FW vivo-extended/FW|JJ). 
  

3.10 Other foreign words

• Liuwei/FW Dihuang/FW Pills/NNS (Chinese medicine)
• bacille/FW Calmette-Guerin/FW vaccine/NN (BCG; French)

3.11 Suffixes

E.g.:

• XXX- and ZZZ-YYY
• XXX ( Abbreviation of XXX ) -YYY

E.g.:

• pre-/JJ and/CC post-transcription/NN
• pre-/JJ and/CC post-transcriptional/JJ
• pre-/RB and/CC post-transcriptionally/RB
• up-/RP and/CC down-regulates/VBZ
• interleukin/NN (/( IL/NN )/) -2/CD
• antigen-dependent/JJ and -independent/JJ
• nuclear/JJ factor/NN (/( NF/NN )/) -kappa/NN B/NN

E.g.:

lipopolysaccharide/NN (/( LPS/NN )/) -induced/JJ TF/NN expression/NN

3.12 Acronyms

E.g.:

TI - Regulation of fas-ligand expression
"fas-ligand" is a hyphenated prenominal word and is tagged as JJ. Thus, "FasL", the abbreviation thereof, should also be tagged as JJ. Acronyms whose original forms cannot be found in the same abstract should by default be tagged as NN.

3.13 NN(P) or NN(P)S

3.13.1 Acronyms and abbreviations of units

Acronyms and abbreviations of units which do not have plural suffixes should be tagged according to different criteria described below.

Acronyms

• The grammatical number of acronyms should be made consistent with the original . forms that are spelt out in the same abstract.

  E.g.:

  protein/NN kinase/NN C/NN (/( PKC/NN )/)

  Acronyms should in principle be tagged consistently within a single abstract. (Hence, the same acronym might be assigned different tags in different abstracts.) Acronyms that are tagged as NNS in other parts of an abstract should also be tagged as NNS in the prenominal position. However,
  • if an acronym whose original form is plural appears both with and without the plural suffix, those with the suffix should be tagged as NNS and those without the suffix should be tagged as NN. Hence, something like the following could happen against the principle above in one and the same abstract:

    
    

    Epstein-Barr viruses ( EBV/NN ) ... EBVs/NNS ...

    Note that, in this case, if the form "EBVs" did not appear, "EBV" would be tagged as NNS throughout the abstract.
  • if the same acronym appears in both positions that are restricted to singular forms and positions that are restricted to plural forms, tokens in each environment should be tagged as NN and NNS, respectively. Those that appear in positions where both singular and plural forms can appear should be tagged as NN. In this case, if an instance of the acronym immediately following the original form enclosed by parentheses doesn't have the plural suffix, it should be tagged as NNS. (The acronym in this position is considered to be an abbreviation of the plural form.)

Abbreviations of units

E.g.:

base pairs ( bp/NN )

3.13.2 Other cases

• The grammatical number of fractions should agree with the number of the numerator (e.g. sites/cell/NNS, fmol/micrograms/NN⁸).
• Words that end in "s" whose singular and plural forms are identical should be tagged as NN by default. If they appear in positions where only plural forms can appear, they should be tagged as NNS.

  E.g.:

  • by means/NN of (Although the WSJ Corpus tag this as NNS, we tag it as . NN.)
  • {a, the} series/NN
  • by all means/NNS

• "pp." which abbreviates for "pages" should be tagged as NNS.

3.14 Slashes

• Expressions in which two words whose independent POSes are either NN or JJ are concatenated and which function as prenominal modifiers should be tagged as JJ. (In such cases, the function of slashes are almost identical to that of hyphens.)
• Other words concatenated by slashes are in general of the same part of speech. Hence, the whole concatenated unit should be assigned the POS common to both components.

3.15 Mistakes in the original text

• Spelling mistakes.

  E.g.:

  @@ Their absence in endometriotic lesions together with increased CD+8 T-cell numbers and increased oestrogen receptor and bcl-2 expression may have significant effects on the development and progression of endometriosis

  "CD+8" in this sentence is a misspelling of "CD8+". Hence, the sentence should be prefixed with "@@". Misspelled words should be tagged as if they were spelled correctly.

  E.g.:

  @@ AB/LS! -/:! Alcoholic/JJ extracts/NNS prepared/VBN? form/IN@@ Arnicae/FW! flos/FW!

  In this case, it can be inferred from the context that "form" is a misspelling of "from". Hence, it should be tagged as "form/IN@@" as above. (Note that this rule goes against the POS-tagging convention stated in the PTB Manual. )
• Agreement mismatch between subject and verb. (The tag of the subject should agree with the grammatical number of the verb and the sentence should be prefixed with "@@".)

  E.g.:

  @@ The/DT results/NN indicates/VBZ that/IN carboranes/NNS are/VBP applicable/JJ as/IN the/DT hydrophobic/JJ moiety/NN of/IN biologically/RB active/JJ molecules/NNS ./.!

  Note, however, that "Degree Expression + Plural Noun" can function as a plural expression and hence should not be prefixed with "@@".

  E.g.:

  ... and a high percentage of infiltrating T cells in rheumatoid arthritis synovium express/VBP receptors for FGF-1 .

• Necessary word dropped.

  E.g.:

  @@ However , no real progress have occurred in diseases such as multiple myeloma , the association of melphalan and prednisone is still the mainstay of the treatment .

  For this sentence to be grammatical, there should be "though" between "However" and "no".

  However, note that gapping is exempt from this generalization. It is a grammatical construction and hence should not be prefixed with "@@".

  E.g.:

  CD38+ cells committed towards this lineage were enriched and a population of these cells treated with dihydroxyvitamin D3 to induce neutrophilmaturation .

  "was" is missing before "treated".
• Note that expressions that are considered to be ungrammatical but that frequently appear in abstracts do not have to be prefixed with "@@". However, obvious grammatical errors should be prefixed with "@@".

  E.g.:

  EC were transfected with either reporter constructs containing the luciferase gene driven either by E-selectin or interleukin (IL)-8 promoters or a synthetic NF-kappaB-dependent promoter .

  In this sentence, NP and PP are coordinated as in "either [PP by N] or [NP N]" but this is not regarded as a mistake.

3.16 Misc

• "ago" in expressions like "several year s ago" should be tagged as RB.
• "irrespective of ..." should be tagged as "irrespective/RB of/IN ..." since it modifies the whole sentence adverbially.
• "post" used in the meaning of "after" should be tagged as IN.
• If you feel uncertain whether or not to put "@@" or"?", put them rather than omit them.
• All instances of "when" should be tagged as WRB. Though the PTB Manual states that it should be tagged as IN if it is used "in the sense of `if' ", we don't adopt this rule since it is often difficult to distinguish cases in which "when" should be tagged as IN from cases in which it should be tagged as WRB.

Notes

1. This is done so for -ed words whose POS tags are almost always VBN in order to reduce the amount of manual POS-fixing.
2. However, for a word that can never be NNS, asterinsks should be left as they are even if the word in question matches the pattern described here (e.g. **NF-kappaB**/NN). If you find asterisks missing from such words, send an email to other annotators and persons concerned.
3. Strictly speaking, not all of the following rules conflict with one another. So, it is not necessary to specify the order among them as rigorously as is done below. For example, the rules 1, 2 and 3 cover distinct cases, which means that the ordering among them is completely irrelevant. In general, the order among them should be something like "[1,2,3] > [4,5] > 6 > 7 > 8". In what follows, however, all the rules are ordered just to maintain the format consistent.
4. This rule overrides the next one. Hence, when "NN XX-YYing" in "NN XX-YYing NN" is a prenominal modifier, "XX-YYing" should be tagged as JJ.
5. Incidentally, "DNA-binding" should be tagged as JJ in vast majority of other cases. E.g.: DNA-binding/JJ {region, sites, protein, complex}
6. Note that not all instances of "I" are roman numerals.
7. When followed by a closing parenthesis like "1)", they should be segmented and tagged like "1/LS )/)".
8. Incidentally, note that although the number of the denominator is plural in this example, which is considered to be a mistake on the part of the author, this is entirely irrelevant to the number of the whole expression.

-ed

-ing

JJ (those marked with * has an adjective entry/entries in LDOCE)

• adhering/JJ cells
• adhering/JJ cellscirculating/JJ levels of cortisol
• circulating/JJ levels of cortisol
• clipping/JJ bandshift assays
• corresponding/JJ NN (all occurrences in WSJ were tagged as JJ)
• cross-linking/JJ CD40
• cutting/JJ edge
• cytoreductive staging/JJ laparotomy
• determining/JJ role
• housekeeping/JJ genes
• impeding/JJ NN
• infiltrating/JJ ductal carcinomas (「浸潤性の」という意味:modifies cancer-related nouns)
• it is tempting/JJ* to ...
• opposing/JJ effects (as this does not mean an effect called "opposing")
• preexisting/JJ*, latent, STAT1-SATAT2 signaling complex
• presenting/JJ HLA molecules
• priming/JJ cultures （meaning "primary"）
• proliferating/JJ state
• cytoreductive staging/JJ laparotomy
• striking/JJ {similarities, differences etc｝/NN
• surrounding/JJ* NN (only before noun)
• it is tempting/JJ* to ...
• underlying/JJ* difference
• unifying/JJ hypothesis
• working/JJ* hypothesis

NN

• non-DNA/JJ binding/NN sequence (to be consistent with "DNA/NN binding/NN sequence")
• blocking/NN factor
• clustering/NN algorithm (method)
• comparative mapping/NN data
• docking/NN sites
• reading/NN frame
• blocking/NN factor
• signaling/NN cascade
• nuclear targeting/NN sequence
• testing/NN data

VBG

• activating/VBG Gi proteins
• activating/VBG region
• multiple activating/VBG agents
• alternating/VBG RCGCRYGCGY consensus
• binding/VBG protein (motif, site)
• blocking/VBG Abs
• budding/VBG yeast
• circulating/VBG human lymphocytes
• coding/VBG region
• crossreacting/VBG? proteins
• deactivating/VBG cytokine
• developing/VBG NN
• differentiating/VBG agent
• differentiating/VBG hematopoietic progenitors
• following/VBG NN （NNにしたがっての意味）
• homodimerizing/VBG cytokine receptors
• internalizing/VBG antibodies
• intervening/VBG NN
• intracellular signaling/VBG agents
• ionizing/VBG radiation
• limiting/VBG element
• rate limiting/VBG
• comparative mapping/NN data
• mediating/VBG receptor-G protein
• menstruating/VBG women (patients)
• multiple activating/VBG agents
• overlapping/VBG endothelium
• proliferating/VBG cells
• resting/VBG monocytes (cells, etc)
• signaling/VBG molecules (agents, proteins, etc)
• smoking/VBG patients
• splicing/VBG NN
• stimulating/VBG T cell blasts
• transforming/VBG growth factor
• ubiquitinylating/VBG enzymes

PREPOS data

Foreign Terms

Sicentific Names of Organisms(treated as Latin)

Other Terms of Latin Origin

Greek Letters

Other terms of foreign origin

Confusing Singular/Plural Forms

Multi-word prepositions and other fixed expressions

Numeric Expressions

Units: treat as nouns

Symbols

Phreses Treated as Adjectives

• For gerunds and paticiples of verbs, refer to the ED-list and ING-list sections.
• Treat as a noun when it is a nominal head.

Abbreviations

Other Adjectives

Verbs

Names with Hyphens: treated as nouns