{\rtf1\ansi\ansicpg1252\uc1 \deff4\deflang1033\deflangfe1033{\fonttbl{\f0\froman\fcharset0\fprq2{\*\panose 02020603050405020304}Times New Roman;}{\f1\fswiss\fcharset0\fprq2{\*\panose 020b0604020202020204}Arial;}
{\f3\froman\fcharset2\fprq2{\*\panose 05050102010706020507}Symbol;}{\f4\froman\fcharset0\fprq2{\*\panose 120060d51200402b1300}TIMES;}{\f5\fswiss\fcharset0\fprq2{\*\panose 020b0604020202020204}Helvetica;}
{\f6\froman\fcharset0\fprq2{\*\panose 00000000000000000000}Courier;}{\f15\fswiss\fcharset0\fprq2{\*\panose 020b0604030504040204}Tahoma;}{\f84\froman\fcharset238\fprq2 Times New Roman CE;}{\f85\froman\fcharset204\fprq2 Times New Roman Cyr;}
{\f87\froman\fcharset161\fprq2 Times New Roman Greek;}{\f88\froman\fcharset162\fprq2 Times New Roman Tur;}{\f89\froman\fcharset186\fprq2 Times New Roman Baltic;}{\f90\fswiss\fcharset238\fprq2 Arial CE;}{\f91\fswiss\fcharset204\fprq2 Arial Cyr;}
{\f93\fswiss\fcharset161\fprq2 Arial Greek;}{\f94\fswiss\fcharset162\fprq2 Arial Tur;}{\f95\fswiss\fcharset186\fprq2 Arial Baltic;}{\f174\fswiss\fcharset238\fprq2 Tahoma CE;}{\f175\fswiss\fcharset204\fprq2 Tahoma Cyr;}
{\f177\fswiss\fcharset161\fprq2 Tahoma Greek;}{\f178\fswiss\fcharset162\fprq2 Tahoma Tur;}{\f179\fswiss\fcharset186\fprq2 Tahoma Baltic;}}{\colortbl;\red0\green0\blue0;\red0\green0\blue255;\red0\green255\blue255;\red0\green255\blue0;
\red255\green0\blue255;\red255\green0\blue0;\red255\green255\blue0;\red255\green255\blue255;\red0\green0\blue128;\red0\green128\blue128;\red0\green128\blue0;\red128\green0\blue128;\red128\green0\blue0;\red128\green128\blue0;\red128\green128\blue128;
\red192\green192\blue192;}{\stylesheet{\qj\fi227\widctlpar\adjustright \f4\fs20\cgrid \snext0 Normal;}{\s1\qj\sa1600\sl-320\slmult0\keep\keepn\pagebb\widctlpar\tx284\hyphpar0\adjustright \b\f4\fs28\cgrid \sbasedon0 \snext0 heading 1;}{
\s2\qj\sb520\sa280\sl-280\slmult0\keep\keepn\widctlpar\tx454\hyphpar0\adjustright \b\f4\fs20\cgrid \sbasedon0 \snext0 heading 2;}{\s3\qj\sb440\sa220\sl-240\slmult0\keep\keepn\widctlpar\tx510\hyphpar0\adjustright \b\f4\fs20\cgrid \sbasedon0 \snext0 
heading 3;}{\s4\qj\sb240\sa60\keepn\widctlpar\ls1\ilvl3\adjustright \b\f1\cgrid \sbasedon0 \snext0 heading 4;}{\s5\qj\sb240\sa60\widctlpar\ls1\ilvl4\adjustright \f1\fs22\cgrid \sbasedon0 \snext0 heading 5;}{
\s6\qj\sb240\sa60\widctlpar\ls1\ilvl5\adjustright \i\f4\fs22\cgrid \sbasedon0 \snext0 heading 6;}{\s7\qj\sb240\sa60\widctlpar\ls1\ilvl6\adjustright \f1\fs20\cgrid \sbasedon0 \snext0 heading 7;}{\s8\qj\sb240\sa60\widctlpar\ls1\ilvl7\adjustright 
\i\f1\fs20\cgrid \sbasedon0 \snext0 heading 8;}{\s9\qj\sb240\sa60\widctlpar\ls1\ilvl8\adjustright \b\i\f1\fs18\cgrid \sbasedon0 \snext0 heading 9;}{\*\cs10 \additive Default Paragraph Font;}{\s15\qj\fi227\widctlpar\tqc\tx4536\tqr\tx9072\adjustright 
\f4\fs20\cgrid \sbasedon0 \snext15 header;}{\s16\qj\fi227\widctlpar\tqc\tx4536\tqr\tx9072\adjustright \f4\fs20\cgrid \sbasedon0 \snext16 footer;}{\*\cs17 \additive \sbasedon10 page number;}{\s18\qc\fi227\sa460\keep\keepn\pagebb\widctlpar
\tx284\hyphpar0\adjustright \b\f4\fs28\cgrid \sbasedon0 \snext19 title;}{\s19\qc\fi227\sa220\widctlpar\adjustright \f4\fs20\cgrid \sbasedon0 \snext20 author;}{\s20\qc\fi227\widctlpar\adjustright \f4\fs18\cgrid \sbasedon0 \snext21 address;}{
\s21\qc\fi227\widctlpar\adjustright \f6\fs18\cgrid \sbasedon0 \snext28 email;}{\s22\qj\sb520\sa280\keep\keepn\widctlpar\tx454\hyphpar0\adjustright \b\f4\cgrid \sbasedon0 \snext29 heading1;}{\s23\qj\sb440\sa220\keep\keepn\widctlpar
\tx510\hyphpar0\adjustright \b\f4\fs20\cgrid \sbasedon0 \snext29 heading2;}{\s24\qj\sb320\keep\keepn\widctlpar\tx284\hyphpar0\adjustright \b\f4\fs20\cgrid \sbasedon0 \snext29 heading3;}{\s25\qc\fi227\li227\sb120\sa120\widctlpar\tx6237\adjustright 
\f4\fs20\cgrid \sbasedon0 \snext0 equation;}{\s26\qj\sb120\sa240\keep\keepn\widctlpar\adjustright \f4\fs18\cgrid \sbasedon0 \snext0 figure legend;}{\s27\qj\sb240\sa120\keep\keepn\widctlpar\adjustright \f4\fs18\lang1031\cgrid \sbasedon0 \snext0 
table title;}{\s28\qj\li567\ri567\sb600\sa120\widctlpar\adjustright \f4\fs18\cgrid \sbasedon29 \snext22 abstract;}{\s29\qj\widctlpar\adjustright \f4\fs20\cgrid \sbasedon0 \snext0 p1a;}{\s30\qj\fi-227\li227\widctlpar\adjustright \f4\fs18\cgrid 
\sbasedon0 \snext30 referenceitem;}{\*\cs31 \additive \fs12\up6\nosupersub \sbasedon10 footnote reference;}{\s32\sa240\widctlpar\tx680\tqr\tx6237\tqr\tx6917\adjustright \f4\fs18\cgrid \sbasedon0 \snext32 Running head - left;}{\s33\qr\sa240\widctlpar
\tx680\tqr\tx6237\tqr\tx6917\adjustright \f4\fs18\cgrid \sbasedon32 \snext33 Running head - right;}{\s34\qj\fi-227\li227\widctlpar\tx227\tx454{\*\pn \pnlvlbody\ilvl11\ls2047\pnrnot0\pnf5\pnstart1\pnindent227\pnhang{\pntxtb ?}}\ls2047\ilvl11\adjustright 
\f4\fs20\cgrid \sbasedon35 \snext34 Bullet Item;}{\s35\qj\fi-227\li227\widctlpar\tx227\tx454{\*\pn \pnlvlbody\ilvl11\ls2047\pnrnot0\pnf5\pnstart1\pnindent227\pnhang{\pntxtb ?}}\ls2047\ilvl11\adjustright \f4\fs20\cgrid \sbasedon0 \snext0 Item;}{
\s36\qj\fi-227\li227\widctlpar\tx227\tx454{\*\pn \pnlvlbody\ilvl11\ls2047\pnrnot0\pndec\pnstart1\pnindent227\pnhang{\pntxta .}}\ls2047\ilvl11\adjustright \f4\fs20\cgrid \sbasedon35 \snext36 Numbered Item;}{\s37\qj\fi-170\li170\sl-220\slmult0\widctlpar
\tx170\adjustright \f4\fs18\cgrid \sbasedon0 \snext37 footnote text;}{\s38\li227\sb120\sa120\widctlpar\tx1361\tx1531\tx1701\tx1871\tx2041\tx2211\tx2381\tx2552\adjustright \f6\fs20\cgrid \sbasedon0 \snext38 programcode;}{\s39\qj\fi-170\li170\widctlpar
\tx170\adjustright \f4\fs18\cgrid \sbasedon0 \snext39 Fu\'dfnotentext.Footnote;}{\s40\qj\fi227\sb120\sa120\widctlpar\adjustright \b\f4\fs20\cgrid \sbasedon0 \snext0 caption;}{\s41\qj\sb320\widctlpar\adjustright \i\f4\fs20\cgrid \sbasedon0 \snext29 
heading4;}{\*\cs42 \additive \ul\cf2 \sbasedon10 Hyperlink;}{\s43\qj\fi227\widctlpar\adjustright \f4\fs20\cgrid \sbasedon0 \snext43 Body Text 2;}{\s44\qj\fi227\widctlpar\adjustright \cbpat9 \f15\fs20\cgrid \sbasedon0 \snext44 Document Map;}{
\s45\qj\li302\ri259\sb20\sa20\keep\nowidctlpar\adjustright \f4\fs22\lang2057\cgrid \sbasedon0 \snext45 Abstract;}{\s46\qj\fi-180\li180\sa60\widctlpar\tx4000\tx10279\adjustright \f4\fs20\lang2057\cgrid \sbasedon0 \snext46 Bibliography;}}{\*\listtable
{\list\listtemplateid-1{\listlevel\levelnfc255\leveljc0\levelfollow0\levelstartat0\levelspace0\levelindent0{\leveltext\'00;}{\levelnumbers;}}{\listlevel\levelnfc255\leveljc0\levelfollow0\levelstartat0\levelspace0\levelindent0{\leveltext
\'00;}{\levelnumbers;}}{\listlevel\levelnfc255\leveljc0\levelfollow0\levelstartat0\levelspace0\levelindent0{\leveltext\'00;}{\levelnumbers;}}{\listlevel\levelnfc0\leveljc0\levelfollow0\levelstartat0\levelold\levelspace0\levelindent0{\leveltext
\'01\'03;}{\levelnumbers\'01;}\f0\fbias0 \s4}{\listlevel\levelnfc0\leveljc0\levelfollow0\levelstartat0\levelold\levelspace0\levelindent0{\leveltext\'01\'04;}{\levelnumbers\'01;}\f0\fbias0 \s5}{\listlevel\levelnfc0\leveljc0\levelfollow0\levelstartat0
\levelold\levelspace0\levelindent0{\leveltext\'01\'05;}{\levelnumbers\'01;}\f0\fbias0 \s6}{\listlevel\levelnfc0\leveljc0\levelfollow0\levelstartat0\levelold\levelspace0\levelindent0{\leveltext\'01\'06;}{\levelnumbers\'01;}\f0\fbias0 \s7}{\listlevel
\levelnfc0\leveljc0\levelfollow0\levelstartat0\levelold\levelspace0\levelindent0{\leveltext\'01\'07;}{\levelnumbers\'01;}\f0\fbias0 \s8}{\listlevel\levelnfc0\leveljc0\levelfollow0\levelstartat0\levelold\levelspace0\levelindent0{\leveltext
\'01\'08;}{\levelnumbers\'01;}\f0\fbias0 \s9}{\listname ;}\listid-5}{\list\listtemplateid-1\listsimple{\listlevel\levelnfc0\leveljc0\levelfollow0\levelstartat0\levelspace0\levelindent0{\leveltext\'01*;}{\levelnumbers;}}{\listname ;}\listid-2}
{\list\listtemplateid68419599\listsimple{\listlevel\levelnfc0\leveljc0\levelfollow0\levelstartat1\levelspace0\levelindent0{\leveltext\'02\'00.;}{\levelnumbers\'01;}\fi-360\li360\jclisttab\tx360 }{\listname ;}\listid86079172}{\list\listtemplateid68419599
\listsimple{\listlevel\levelnfc0\leveljc0\levelfollow0\levelstartat1\levelspace0\levelindent0{\leveltext\'02\'00.;}{\levelnumbers\'01;}\fi-360\li360\jclisttab\tx360 }{\listname ;}\listid308677128}}{\*\listoverridetable{\listoverride\listid-5
\listoverridecount0\ls1}{\listoverride\listid-2\listoverridecount1{\lfolevel\listoverrideformat{\listlevel\levelnfc23\leveljc0\levelfollow0\levelstartat1\levelold\levelspace0\levelindent227{\leveltext\'01\u-4051 ?;}{\levelnumbers;}\f3\fbias0 
\fi-227\li227 }}\ls2}{\listoverride\listid-2\listoverridecount1{\lfolevel\listoverrideformat{\listlevel\levelnfc23\leveljc0\levelfollow0\levelstartat0\levelold\levelspace0\levelindent360{\leveltext\'01\u-3913 ?;}{\levelnumbers;}\f3\fbias0 \fi-360\li720 }}
\ls3}{\listoverride\listid308677128\listoverridecount0\ls4}{\listoverride\listid86079172\listoverridecount0\ls5}}{\info{\title Lecture Notes in Computer Science:}{\author foobar}{\operator Diana Santos}{\creatim\yr2003\mo2\dy7\hr12\min55}
{\revtim\yr2003\mo3\dy24\hr13\min34}{\printim\yr2003\mo3\dy24\hr12\min45}{\version7}{\edmins118}{\nofpages8}{\nofwords3203}{\nofchars18261}{\*\company Springer Verlag}{\nofcharsws0}{\vern89}}
\paperw11907\paperh16840\margl2495\margr2495\margt2948\margb2948 \deftab708\widowctrl\ftnbj\aenddoc\hyphhotz425\makebackup\margmirror\lytprtmet\hyphcaps0\hyphauto1\formshade\viewkind4\viewscale120\pgbrdrhead\pgbrdrfoot \fet0\sectd 
\psz9\linex0\headery2381\footery1389\colsx709\sectdefaultcl {\*\pnseclvl1\pnucrm\pnstart1\pnindent720\pnhang{\pntxta .}}{\*\pnseclvl2\pnucltr\pnstart1\pnindent720\pnhang{\pntxta .}}{\*\pnseclvl3\pndec\pnstart1\pnindent720\pnhang{\pntxta .}}{\*\pnseclvl4
\pnlcltr\pnstart1\pnindent720\pnhang{\pntxta )}}{\*\pnseclvl5\pndec\pnstart1\pnindent720\pnhang{\pntxtb (}{\pntxta )}}{\*\pnseclvl6\pnlcltr\pnstart1\pnindent720\pnhang{\pntxtb (}{\pntxta )}}{\*\pnseclvl7\pnlcrm\pnstart1\pnindent720\pnhang{\pntxtb (}
{\pntxta )}}{\*\pnseclvl8\pnlcltr\pnstart1\pnindent720\pnhang{\pntxtb (}{\pntxta )}}{\*\pnseclvl9\pnlcrm\pnstart1\pnindent720\pnhang{\pntxtb (}{\pntxta )}}\pard\plain \s18\qc\fi227\sa460\keep\keepn\pagebb\widctlpar\tx284\hyphpar0\adjustright 
\b\f4\fs28\cgrid {Timber! Issues in treebank building and use
\par }\pard\plain \s19\qc\fi227\sa220\widctlpar\adjustright \f4\fs20\cgrid {Diana Santos}{\fs14\up6 1}{
\par }\pard\plain \s20\qc\fi227\widctlpar\adjustright \f4\fs18\cgrid {\fs12\up6 1}{ Linguateca, SINTEF Telecom & Informatics, Pb 1124 Blindern,\line 0314 Oslo, Norway
\par }\pard\plain \s21\qc\fi227\widctlpar\adjustright \f6\fs18\cgrid {\cs42\ul\cf2 Diana.Santos@sintef.no}{
\par }\pard\plain \s28\qj\li567\ri567\sb600\sa120\widctlpar\adjustright \f4\fs18\cgrid {\b Abstract.}{ }{\lang2070 We discuss several treebank conceptions in the literature and show that their require
ments may be incompatible, describing then the options taken in the construction of a Portuguese treebank, in what concerns human vs. aut}{\lang2070 o}{\lang2070 matic intervention. Use cases are then listed in connection with a Web search tool (}{
\i\lang2070 \'c1guia}{\lang2070 ), whose philosophy and implementation is presented.
\par }\pard\plain \s22\qj\sb520\sa280\keep\keepn\widctlpar\tx454\hyphpar0\outlinelevel0\adjustright \b\f4\cgrid {1\tab Introduction
\par }\pard\plain \qj\fi227\widctlpar\adjustright \f4\fs20\cgrid {\lang2070 Treebank building has become fashionable lately with the number of treebank pr}{\lang2070 o}{\lang2070 
jects growing exponentially. However, there are quite different ways to conceive both the end result and the way to go about achieving it.
\par As far as treebank purpose is concerned, one can identify at least the following di}{\lang2070 f}{\lang2070 ferent views (an example of each is provided with no claim for exhaustiveness):
\par 1. a treebank as a resource for the building of automatic processing tools [1]
\par 2. a treebank is an evaluation resource to compare the performance of different pa}{\lang2070 r}{\lang2070 sers [2]
\par 3. a treebank is a linguistic resource to fix and display the syntactic analysis of complex text (and can consequently be used for teaching purposes) [3]
\par 4. a treebank is a proof of the qualities of a given theory}{\cs31\fs12\up6\lang2070 \chftn {\footnote \pard\plain \s37\qj\fi-170\li170\sl-220\slmult0\widctlpar\tx170\adjustright \f4\fs18\cgrid {\cs31\fs12\up6 \chftn }{
 This is rarely stated but it often constitutes an additional motive to engage in treebank buil}{d}{ing.}}}{\lang2070  
\par Even though most papers on treebanks so far declare that they expect it to be used for (almost) all these purposes, a closer analysis show that the requirements to achieve these different goals are incompatible or, at least, difficult to harmonize.
\par For example, if you want to train computer programs on the treebank, you\rquote d better only revise and clean information about which there is some understanding on how to program or achieve. In other words, information added by a human drawn from sou}{
\lang2070 r}{\lang2070 ces such as world knowledge or cognitive processing difficulties, as well as the result of complex inferences based on a distant context are not, in general, reproducible automatically and are therefore of no interest for goal 1.

\par In fact, desirable features for a treebank type 1 are: consistency, few information pieces and enough occurrences of each feature (so that systems have enough examples from which to learn).
\par On the other hand, if one wants to create a gold standard for ensuing evaluation e}{\lang2070 n}{\lang2070 deavours, it is possible that one chooses not to annotate, or not to decide in cases wh}{\lang2070 e}{\lang2070 
re consensus was not reached. The result may not be consistent or complete, but it is empirically adequate.
\par }{\lang2057 If one wants to use a treebank for linguistic investigation, one would value most of all the information that only linguists could add, and actually almost \ldblquote despise\rdblquote  the sort of low-
level information that satisfaction of goal 1 would require (like correct mo}{\lang2057 r}{\lang2057 phological information). Consistency would 
be a platonic goal, but naturalness of the annotation and relevance to linguistic concerns would be features of such a treebank type 3.
\par Finally, a treebank type 4 should maximize diversity (although keeping consi}{\lang2057 s}{\lang2057 tency) in order to prove the expressiveness of the theory and therefore would again fail to be useful for goal 1.
\par Our treebank project, }{\i\lang2057 Floresta Sint\'e1(c)tica}{\lang2057  [4], aimed (eventually) at building a type 3 treebank, given that we had an underlying symbolic parser which provided a lot of information an
d it was unrealistic to expect that a parser could be trained to learn it all. Reducing it would be a bold decision, which was not taken.
\par }\pard\plain \s22\qj\sb520\sa280\keep\keepn\widctlpar\tx454\hyphpar0\adjustright \b\f4\cgrid {2\tab Annotation schemes
\par }\pard\plain \s43\qj\fi227\widctlpar\adjustright \f4\fs20\cgrid {\lang2057 Wilson et al. [5] describe a set of desiderata for an annotation scheme where they emphasize that it should reflect distinctions a human could be expected to reliably annotate (
\ldblquote naturalness\rdblquote ). It is easy to find huge numbers of information tags that are not easy to annotate reliably (even though they may be used liberally by parsers); it is also the c
ase that many of the easy to annotate categories for humans are, so far, never even attempted automatically.
\par }\pard\plain \s23\qj\sb440\sa220\keep\keepn\widctlpar\tx510\hyphpar0\adjustright \b\f4\fs20\cgrid {2.1\tab Can our treebank type 3 be turned into an evaluation treebank (type 2)?
\par }\pard\plain \qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 How to create a treebank that allows one to actually evaluate different parsers w}{\lang2070 i}{\lang2070 
thout forcing the linguistic view of the present treebank? Although we, as creators, might wish that it took the same role as the Penn Treebank [6] for English, used as a de facto standard, we are fully convinced of the need and advantage of coo
peratively agreeing on a standard.
\par We believe that the present treebank can be used for experimentation and evalu}{\lang2070 a}{\lang2070 tion, and to make problems and disagreements explicit, but that one should try to build from scratch (or from a much stricter set of rules and
 using as point of departure the present treebank) a real evaluation resource that allows one to test given aspects of syntactic parsers for Portuguese, probably following Gaizauskas et al.\rquote 
s proposal [7] for creating evaluation resources quickly, and using some manual analysis as in [8].
\par We are, in any case, convinced that it is totally unrealistic to expect that one can list parsers\rquote  outputs and try to harmonize or agree on the meaning of the different labels. This was already an enormous task for a field
 as (comparatively) simple as Portuguese morphological analysis, for which an unexpected high degree of disagreement has been reported [9,10]. It is also enough to browse several different Portuguese grammar books to see that they verse about different su
bjects. Incidentally, it is also quite rare that they define their primitives.
\par }\pard\plain \s24\qj\sb320\keep\keepn\widctlpar\tx284\hyphpar0\adjustright \b\f4\fs20\cgrid {\lang2070 2.1.1 \tab Decisions as to the process
\par }\pard\plain \qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 Let us give a concrete example of one of the many things that are far from trivial: The underlying parser \endash  thoroughly described in [11] \endash  assi
gns the two following syntactic categories to noun phrases attached to noun phrases: N<PRED (predicative adject) and APP ((adnominal) apposition), exemplified respectively as }{\i\lang2070 Jer\'f4nimo, }{\b\i\lang2070 um grande cacique}{\i\lang2070 
, temia ningu\'e9m}{\lang2070  and }{\i\lang2070 O grande cacique}{\b\i\lang2070 , Jer\'f4nimo}{\i\lang2070 , conhecia o seu pa\'eds como mais ningu\'e9m}{\lang2070 . The definitions in the treebank documentation follow: 
\par }{\fs16\lang2070 APP: The prototypical apposition is a name or definite np, identifying the np-head it postmodifies: "Jer\'f3nimo, o grande cacique" or "o seu advogado, Marco da Silva". 
\par N<PRED: The prototypical postnominal predicative is an adjective, attributive participle or indefinite np, predicating something about the np-head it postmodifies, typically with the semantic relation of 'IS' (=): "Jorge Gomes, funcion\'e1
rio" or "Jorge Gomes, contente com a vida".
\par }{\lang2070 It has proved, no matter the many heuristics or rules of thumb proposed}{\cs31\fs12\up6\lang2070 \chftn {\footnote \pard\plain \s37\qj\fi-170\li170\sl-220\slmult0\widctlpar\tx170\adjustright \f4\fs18\cgrid {\cs31\fs12\up6 \chftn }{ 
Such as: when an abbreviation follows what it is an abbreviation for, tag it @APP: }{\i Partido da Terra }{\b\i (PT)}{; APP implies an identity relation, while N<PRED adds information, ...}}}{\lang2070 , an extr}{\lang2070 e}{\lang2070 
mely difficult decision to be done in practice, when one leaves the idealized landscape and comes to real utterances. Time and again there was uncertainty about which cla}{\lang2070 s}{\lang2070 sification to assign. Examples are:
\par }{\i\lang2070 No final do jogo, adeptos do Sporting lan\'e7am garras e pedras para a tribuna de honra, onde estavam Manuela Ferreira Leite, }{\b\i\lang2070 ministra da Educa\'e7\'e3o}{\i\lang2070 , e V\'edtor Vasques, }{\b\i\lang2070 presidente da FPF}{
\i\lang2070 .
\par Na mesma zona em que foi encontrado o templo, }{\b\i\lang2070 a Alc\'e1\'e7ova}{\i\lang2070 , a caminho das Portas do Sol, foram ainda descobertas cisternas romanas que est\'e3o tamb\'e9m a ser objecto de escava\'e7\'f5es e estudos arqueol\'f3gicos.

\par }{\lang2070 Several solutions about how to proceed concerning the assignment of these labels have been proposed, each of them showing, in fact, different conceptions of what a treebank should be for.
\par 1. mark/revise the clear cases and leave the parser\rquote s output when no clear opinion
\par 2. create a new non-committal label (let us call it here }{\b\lang2070 npstack}{\lang2070 ) and
\par \tab a. transform all cases of either label into it, or
\par \tab b. use it only for the unclear cases
\par Even though no final decision was (so far) taken, this micro-controversy allows us to illustrate the consequences of each option with respect to the treebank goals me}{\lang2070 n}{\lang2070 
tioned in the beginning of the present paper: The first option was aimed at improving the parser, so that it agreed with human reasoning when humans had something to say. The result would probably not be consistent, and definitely not reflecting human pe}
{\lang2070 r}{\lang2070 formance, but was obviously ideal for parser improvement.
\par The second one was, on the contrary, aimed at describing human interpretation (and not a parser\rquote s). Option a) had the goal of making the task of building (and cons}{\lang2070 e}{\lang2070 quently revising) the tre
ebank simpler, taking implicitly the view that this is probably not a human task \endash  when we see two NPs following each other, it is not relevant to understand whether the second is APP or N<PRED. 
\par }\pard\plain \s43\qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 Option b) aimed at pushing the limits of what is encoded in the treebank, to all d}{\lang2070 e}{\lang2070 
cisions a human being could possibly make. So, if in some cases a person can reliably do some distinction, encode it, leaving the rest also encoded as \ldblquote not possible to dec}{\lang2070 i}{\lang2070 de\rdblquote 
, paving the way for a more thorough overview of what can be relevant in interpr}{\lang2070 e}{\lang2070 
ting Portuguese text. The lurking assumption here is that the categories N<PRED and APP do have some relevance for Portuguese grammar, assumption supported by its being used by the parser and mentioned in several traditional grammars.
\par }\pard\plain \s24\qj\sb320\keep\keepn\widctlpar\tx284\hyphpar0\adjustright \b\f4\fs20\cgrid {\lang2070 2.1.2 \tab Decisions as to the encoding
\par }\pard\plain \qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 In [12] it was argued that evaluation of parsed corpora has to take into account at which level a given phenomenon was (or not) represented. In particular, it was pr}{\lang2070 o}{
\lang2070 bably irrelevant to assign right or wrong to PoS classification of }{\i\lang2070 clara}{\lang2070  in }{\i\lang2070 clara e s}{\i\lang2070 u}{\i\lang2070 cintamente}{\lang2070  [13], provided this is recognized as an adverbial phrase.
\par Also, it is even probably wrong to assign the gender and number features masculine plural to }{\i\lang2070 surpresa}{\lang2070  in }{\i\lang2070 presentes surpresa}{\lang2070  although it behaves rather like a common masculine plural adjective like }{
\i\lang2070 caros}{\lang2070  (cf. }{\i\lang2070 Estes presentes foram surpresa}{\lang2070 , }{\i\lang2070 os presentes surpresa est\'e3o no canto}{\lang2070 , }{\i\lang2070 acho estes presentes muito pouco surpresa!}{\lang2070 
) Similarly, when we have a fixed expression like }{\i\lang2070 pele vermelha}{\lang2070 , as in }{\i\lang2070 o chefe pele vermelha bocejou}{\lang2070 
, we can assign to it, in addition to internal features, external features. These two sets of features may or may not agree. So, just as the question of whether }{\i\lang2070 que}{\lang2070 
 is a subject or an object has to be stated relative to the clause in consideration (relative clause or main clause), the question whether }{\i\lang2070 pele}{\lang2070 
 is a noun or an adjective depends on which context: in the noun phrase above or below?}{\cs31\fs12\up6\lang2070 \chftn {\footnote \pard\plain \s37\qj\fi-170\li170\sl-220\slmult0\widctlpar\tx170\adjustright \f4\fs18\cgrid {\cs31\fs12\up6 \chftn }{
 In the present discussion, we are assuming a dependential framework where features are a}{s}{
signed to words (and functional roles are assigned to head words). The need for upwards and downwards marking remains in a more populated phrase structure formalism, we would just have to say "the clause headed by }{\i que}{," or "the phrase headed by }{
\i pele}{".}}}{\lang2070  What is relevant is that }{\i\lang2070 pele}{\lang2070  is a feminine noun for the lower NP (and so requires the adjective }{\i\lang2070 vermelho}{\lang2070  to agree) but behaves as a masculine a}{\lang2070 d}{\lang2070 
jectival phrase (or adjective), in the sentence above.
\par Conflicts may arise when a given lexical item is subject to conflicting requirements due to the different roles it plays, and this may actually even bring changes to the whole language system. For example, if }{\i\lang2070 onde }{\lang2070 vs. }{
\i\lang2070 aonde}{\lang2070  should be selected according to whether the verb describes a movement to some place or not, the two sentences }{\i\lang2070 Vi aonde ele foi}{\lang2070  (\lquote I saw to-where he went\rquote ) and }{\i\lang2070 
Fui onde ele se escondeu}{\lang2070  (\lquote I went where he hid himself\rquote ) are both suboptimal since the two verbs (}{\i\lang2070 ver}{\lang2070  and }{\i\lang2070 ir}{\lang2070  in the first sente}{\lang2070 n}{\lang2070 ce, }{\i\lang2070 ir}{
\lang2070  and }{\i\lang2070 esconder}{\lang2070  in the second) have different features, and therefore requ}{\lang2070 i}{\lang2070 rements.
\par Contrary to a common view that parsed corpora should use the same information as lexicons, we believe that the interest of annotating corpora is precisely to investigate how language works and find out what }{\ul\lang2070 cannot}{\lang2070 
 be predicted from the lexicon, as in }{\i\lang2070 surpresa}{\lang2070  above.
\par }\pard\plain \s22\qj\sb520\sa280\keep\keepn\widctlpar\tx454\hyphpar0\adjustright \b\f4\cgrid {3\tab \'c1guia
\par }\pard\plain \s43\qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 Let us present a Web query tool that has been designed w
ith two considerations in mind: 1. to furnish a higher level query language (in the sense of being as much as possible separate from the encoding realities and actual treebank syntax); 2. to be based on a powerful general purpose corpus system (the IMS CW
B) instead of writing from scratch a particular treebank specific query system.
\par }\pard\plain \qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 This tool is available on the Web (http://www.linguateca.pt/Floresta/) together with a guided tour that tries to give a feeling of the sort of possible queries \endash 
 as high level as possible.
\par }{\i\lang2070 \'c1guia}{\lang2070 \rquote s more radical (or unusual) feature is that its output is simple text, although the whole treebank is publicly available in its two internal coding formats, and ther}{\lang2070 e}{\lang2070 
fore users can, if they want, see and use the tree structure at will. The basis for this feature is that we believe that a treebank user is not (or should not be) primarily co}{\lang2070 n}{\lang2070 
cerned with trees, but with the information conveyed by these trees, in order to get at text, to get at language (which comes in the format of words in the written medium).
\par In addition, we are not yet sure about which are the most interesting questions users really want to ask a treebank. Therefore, we implemented also an open window where people can input questions in natural language and we help them to
 formulate their questions, with the proviso they are answerable by the actual treebank.
\par }\pard\plain \s23\qj\sb440\sa220\keep\keepn\widctlpar\tx510\hyphpar0\adjustright \b\f4\fs20\cgrid {3.1\tab Kinds of queries
\par }\pard\plain \qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 We can distinguish the following kinds of primary uses for a query tool for people (not for programs): }{The user wants }{\ul quantitative information about the treebank}{
, such as: What kind of clauses are most frequent? What kind of syntactic objects (phrases) have the function "question", and in what relative weight? What is the most frequent verb in each kind of clause? What is the most common function of
 a finite clause? In how many cases do adverbs occur in relative clauses?
\par The user wants to }{\ul inspect some combinations or categories}{ a little better \endash  because s/he suspects they are wrongly assigned, or because they contradict her/his own beliefs about the 
language. Some (random) examples: How often can crosscategorial co}{n}{junctions be found? Are there subject complements with relative clauses?
\par The user may simply want to }{\ul look for specific examples of special cases}{, related to his or her field of interest: Fi
nd clauses including an adverb as immediate constituent; find noun phrases including relative clauses in which the pronoun has the subject (or object, or dative) role; find finite clauses starting with the verb, etc.
\par  The user may also want to }{\ul look at the underlying generative grammar}{, according to the examples atested in the treebank: what is the generative grammar of a noun phrase? What is the generation grammar of a particular function? 
\par Or the user may be more interested in the lexicon, and want to }{\ul determine the grammatical properties of a lexical item}{
: what is the valency grammar of a particular lexical item (verb, preposition)? Given a particular class of adverbs, in which patterns do they occur? When a given lexical item occurs as premodifier of a phr
ase, which functions does this phrase typically show?
\par Above, we showed a variety of different questions which could be answered by a }{\ul single}{ query with \'c1guia. There is obviously no limit to the complexity of the intera}{c}{tion an experienced user may have with
 the treebank! We list here other questions that include more than one query but should not be too complicated to answer: What is the deepest embedding? (Find finite clauses under finite clauses.) How many prepositional phrases are not directly attached t
o the preceding phrase? How many noun phrases exhibit a potential attachment ambiguity?
\par Still, other metalinguistic questions, at the moment not catered for by \'c1guia, but encoded in the treebank, can be asnwered: Which sentences were considered ambig}{u}{ous in
 the treebank? Which utterances required world knowledge for disambiguation? (see examples in [14]). Which clauses involve ellipsis or required insertion of add}{i}{tional material in order to be parsed and represented by the human team?
\par }\pard\plain \s23\qj\sb440\sa220\keep\keepn\widctlpar\tx510\hyphpar0\adjustright \b\f4\fs20\cgrid {3.2\tab Use of IMS CWB
\par }\pard\plain \qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 The use of the underlying IMS CWB [15-17] is an obviously sound engineering d}{\lang2070 e}{\lang2070 
cision, since it offers a well developed and tested set of capabilities, a powerful query language and several utilities. In addition, we believe that there should be, at least from
 a user point of view, a smooth transition between POS-tagged and annotated corpora, and the fact that the codification of the latter may pose complex problems to the language engineer should be transparent to the user.
\par The way we used the IMS CWB was stra
ightforward but somehow imaginative: we created several different physical corpora from the manually edited output, that code the treebank in different ways. Depending on the query, the right corpus is used. This is, however, perfectly transparent for the
 user, who can only distinguish between the manually revised part (}{\i\lang2070 Bosque}{\lang2070 , the treebank proper) and the larger automatically pr}{\lang2070 o}{\lang2070 duced part (}{\i\lang2070 Floresta Virgem}{\lang2070 , \ldblquote 
the treebank to be\rdblquote ). 
\par For example, we present an extract of one of the corpora in figure 1, having
 words as terminals and phrases as structural attributes, and therefore appropriate to look for words inside phrases, while the corpus of figure 2 has phrases as terminals and words as attributes.
\par }\pard \fi227\nowidctlpar\adjustright {\f6\fs16\lang2070 <u C22-2>
\par <s>
\par <fcl0 STA>
\par \ldblquote        \ldblquote        pont    0       1
\par <fcl1 ADVL>
\par Se      se      SUB:conj-s      0       2
\par <vp2 P>
\par for     ser     AUX:v-fin       FUT_3S_SUBJ     3
\par firmado firmar  MV:v-pcp        M_S     3
\par </vp2>
\par </fcl1>
\par ,       ,       pont    0       1
\par ningu\'e9m ningu\'e9m SUBJ:pron-indp  M_S     1
\par ficar\'e1  ficar   P:v-fin FUT_3S_IND      1
\par <ap1 SC>
\par mais    mais    >A:adv  <quant> 2
\par contente        contente        H:adj   M_S     2
\par <acl2 KOMP<>
\par do_que  do_que  COM:conj-s      0       3
\par n\'f3s     n\'f3s     SUBJ:pron-pers  M/F_1P_NOM/PIV  3
\par </acl2>
\par </ap1>
\par .       .       pont    0       1
\par </fcl0>
\par }\pard\plain \s26\qc\sb120\sa240\keep\keepn\widctlpar\adjustright \f4\fs18\cgrid {\lang2070 Figure 1: One of the views of the treebank encoded in the IMS-CWB
\par }\pard\plain \qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\f6\fs16\lang2070 <u C22-2>
\par vp    P     'v-fin v-pcp '  'AUX MV '      "for firmado "  2
\par fcl   ADVL  'conj-s vp2 '   'SUB P '    "Se for firmado "       3
\par acl   KOMP< 'conj-s pron-pers '   'COM SUBJ ' "do_que n\'f3s "   2
\par ap    SC    'adv adj acl2 ' '>A H KOMP<' "mais contente do_que n\'f3s" 4
\par fcl  STA   'fcl1 pron-indp v-fin ap1 '  'ADVL SUBJ P SC '  "\ldblquote  Se for firmado , ningu\'e9m ficar\'e1 mais contente do_que n\'f3s . " 12
\par }\pard\plain \s26\qc\sb120\sa240\keep\keepn\widctlpar\adjustright \f4\fs18\cgrid {\lang2070 Figure 2: Another view of the treebank encoded in the IMS-CWB
\par }\pard\plain \qj\fi227\nowidctlpar\adjustright \f4\fs20\cgrid {\lang2070 
\par While it is outside the scope of the present paper to dwell on technicalities, this small section should be read as a plea for using already existing powerful tools for dealing with large amounts of linguisticall
y analysed text, instead of reinventing the wheel and create new treebank search tools from scratch, as was e.g. done in the TIGER project [18].
\par We conclude the present paper asking everyone interested in Portuguese syntax to look at }{\i\lang2070 Floresta Sint\'e1(c)tica}{\lang2070  and try out }{\i\lang2070 \'c1guia}{\lang2070  for the questions they are more int}{\lang2070 e}{\lang2070 
rested in, so that we can have a representative idea of the shortcomings and the main user needs, and may be able to develop a tool that can be generally used, also later on, for different treebanks for Portuguese (and even other languages,
 if the concept turns out to be pertinent).
\par }\pard\plain \s22\qj\sb520\sa280\keep\keepn\widctlpar\tx454\hyphpar0\adjustright \b\f4\cgrid {References
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 1.\tab}}\pard\plain \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright \f4\fs18\cgrid {
Marcus, Mitchell, Kim, Grace, Marcinkiewicz, Mary Ann, MacIntyre, Robert, Bies, Ann, Ferguson, Mark, Katz, Karen, Schasberger, Britta: The Penn treebank: Annotating pred}{i}{
cate argument structure. In: Proceedings of the 1994 Human Language Technology Wor}{k}{shop (ARPA) (1994) 110\emdash 115.
\par {\pntext\pard\plain\s30 \f4\fs18\lang2070\cgrid \hich\af4\dbch\af0\loch\f4 2.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {\lang2070 
Xia, Fei, Palmer, Martha, Xue, Nianwen, Okurowski, Mary Ellen, Kovarik, John, Chiou, Fu-dong, Huang, Shizhe, Kroch, Tony, Marcus, Mitch: Developing Guidelines and Ens}{\lang2070 u}{\lang2070 
ring Consistency for Chinese Text Annotation. In: Gavriladou, M. et al. (eds.), Proceedings of LREC 2000}{\i\lang2070  }{\lang2070 (2000) 3\emdash 10.}{
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 3.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {Skut, Wojciech, Brants, T
horsten, Krenn, Brigitte, Uszkoreit, Hans: A Linguistically Inte}{r}{preted Corpus of German Newspaper Text. In: Rubio, A. et al. (eds.), Proceedings of LREC 1998 (1998) 705\emdash 711}{\lang2070 
\par {\pntext\pard\plain\s30 \f4\fs18\lang2070\cgrid \hich\af4\dbch\af0\loch\f4 4.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {\lang2070 
Afonso, Susana, Bick, Eckhard, Haber, Renato, Santos, Diana: "Floresta sint\'e1(c)tica": a treebank for Portuguese. In: Rodr\'edguez, M.G., Araujo, C.P.S. (eds.): }{\i\lang2070 Proceedings of LREC 2002 }{\lang2070 (2002), 1698\emdash 1703
\par {\pntext\pard\plain\s30 \f4\fs18 \hich\af4\dbch\af0\loch\f4 5.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {\cgrid0 
Wilson, G., Mani, I., Sundheim, B., Ferro, L.: A multilingual approach to annotating and extracting temporal information. In: Proceedings of the Worskhop for Temporal and Sp}{\cgrid0 a}{\cgrid0 tial Information Processing (Toulouse, July 7}{\super\cgrid0 
th}{\cgrid0  2001) (2001) 81\emdash 87
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 6.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {
Marcus, Mitchell P., Santorini, Beatrice, Marcinkiewicz, Mary Ann: Building a large Annotated Corpus of English: The Penn Treebank. }{\i Computational Linguistics}{, }{\b 19}{ (1993) 313\emdash 330
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 7.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {
Gaizauskas, R., Hepple, M., Huyck, C. Modifying Existing Annotated Corpora for General Comparative Evaluation of Parsing. In: Workshop on Evaluation of Parsing Systems, at the LREC'98 (1998)
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 8.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {
Carroll, John, Minnen, Guido, Briscoe, Ted: Corpus annotation for Parser Evaluation. I}{\lang2070 n: Uszkoreit, H. et al (eds.): Proceedings of LINC-99: Linguistically Interpreted Corpora, EACL (Bergen, 12 June 1999) (1999) }{35}{\lang2070 \emdash }{41

\par {\pntext\pard\plain\s30 \f4\fs18\lang2070\cgrid \hich\af4\dbch\af0\loch\f4 9.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {\lang2070 
Santos, Diana, Rocha, Paulo: AvalON: uma iniciativa de avalia\'e7\'e3o conjunta para o port}{\lang2070 u}{\lang2070 gu\'eas. In: Actas do XVIII Encontro da Associa\'e7\'e3o Portuguesa de Lingu\'edstica (Porto, 2-4 de Outubro de 2002) (2003)
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 10.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {Santos, Diana, Costa, Lu}{
\lang2070 \'eds, }{Rocha, Paulo: Cooperatively evaluating Portuguese morpho}{l}{ogy. In: this volume (2003)
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 11.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {Bick, Eckhard: The Pars
ing System "Palavras": Automatic Grammatical Analysis of Po}{r}{tuguese in a Constraint Grammar Framework. Aarhus University Press (2000)
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 12.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {
Santos, Diana, Gasperin, Caroline: }{\f0\lang2070\cgrid0 Evaluation of parsed corpora: experiments in user-transparent and user-visible evaluation. In}{\lang2070  Rodr\'edguez, M.G.; Araujo, C.P.S. (eds.): Pr}{\lang2070 o}{\lang2070 ceedings of LREC 2002}
{\i\lang2070  }{\lang2070 (2002) 597\emdash 604}{
\par {\pntext\pard\plain\s30 \f4\fs18\lang2070\cgrid \hich\af4\dbch\af0\loch\f4 13.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {\lang2070 
Afonso, Susana: Clara e sucintamente: Um estudo em corpus sobre a coordena\'e7\'e3o de adv\'e9rbios em \endash mente. In: }{\i\lang2070 Actas do XVIII Encontro da Associa\'e7\'e3o Portuguesa de Lingu\'ed}{\i\lang2070 s}{\i\lang2070 tica}{\lang2070 
 (Porto, 2-4 de Outubro de 2002) (2003)}{
\par {\pntext\pard\plain\s30 \f4\fs18\lang2070\cgrid \hich\af4\dbch\af0\loch\f4 14.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {\lang2070 
Afonso, Susana, Bick, Eckhard, Haber, Renato, Santos, Diana: Floresta sint\'e1(c)tica: um treebank para o portugu\'eas. In: Gon\'e7alves, Anabela, Correia, Clara Nunes (eds.): }{\i\lang2070 Actas do XVII Encontro da Associa\'e7\'e3o Portuguesa de Lingu
\'edstica }{\lang2070 (Lisboa, 2-4 de Outubro de 2001) (2002) 533\emdash 545}{
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 15.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {
Christ, Oliver: A modular and flexible architecture for an integrated corpus query system. In: }{\i Proceedings of COMPLEX'94: 3rd Conference on Computational Lexicography and Text Research}{ (1994) 23\emdash 32}{\lang2070 
\par {\pntext\pard\plain\s30 \f4\fs18\lang2070\cgrid \hich\af4\dbch\af0\loch\f4 16.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {\lang2070 
Evert, Stefan: }{CQP Query Language Tutorial. IMS Stuttgart, 13 Out 2001}{\lang2070 
\par {\pntext\pard\plain\s30 \f4\fs18\lang1044 \hich\af4\dbch\af0\loch\f4 17.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {\lang1044\cgrid0 
Evert, Stefan; Kermes, Hannah: Annotation, storage, and retrieval of mildly recursive structures. In: Proceedings of the Workshop on Shallow Processing of Large Corpora (SProLaC 2003) (2003)}{\lang2070 
\par {\pntext\pard\plain\s30 \f4\fs18\cgrid \hich\af4\dbch\af0\loch\f4 18.\tab}}\pard \s30\qj\fi-360\li360\widctlpar\jclisttab\tx360{\*\pn \pnlvlbody\ilvl0\ls5\pnrnot0\pndec\pnstart1\pnindent360\pnhang{\pntxta .}}\ls5\adjustright {K\'f6
nig, Esther, Lezius, Wolfgang: A description language for syntactically annotated co}{r}{pora. In: }{\i Proceedings of COLING 2000}{ (2000) 1056\emdash 1060}{\cgrid0  
\par }}