ANNIS2 User Guide - Version 2.1.8 - Hu
Contents
1. For negated tokens word forms use the reserved attribute tok For example lemma sein or tok ist Metadata can also be negated similarly lemma Sein amp meta Genre Sport To only find finite forms of this verb in PCC2 use the part of speech pos annotation concurrently and specify that both the lemma and pos should apply to the same element lemma sein amp pos VAFIN amp 1 _ _ 2 The expression 1 _ _ 2 uses the span identity operator to specify that the first annotation and the second annotation apply to exactly the same position in the corpus Annotations can also apply to longer spans than a single token for example in PCC2 the annotation Inf Stat signifies the information structure status of a discourse referent This annotation can also apply to phrases longer than one token The following query finds spans containing new discourse referents not previously mentioned in the text exmaralda Inf Stat new If the corpus contains no more than one annotation type named Inf Stat the optional namespace in this case exmaralda may be dropped if there are multiple annotations with the same name but different namespaces dropping the namespace will find all of those annotations In order to view the span of tokens to which this annotation applies enter the and click on Show Result then open the exmaralda annotation level to view the grid containing the span Further operators can test the relationsh2. INK NK Mit seinem Tor zum 1 0 Wi Note that since the context is set to a number of tokens left and right of the search term the tree for the whole sentence may not be retrieved To do this you may want to specifically search for the sentence dominating the PP To do so specify the sentence in another element and use the indirect dominance gt operator cat S amp cat PP amp pos NE amp 1 gt 2 amp 2 gt 3 If the annotations in the corpus support it you may also look for edge labels Using the following query will find all adjunct modifiers of a VP dominated by the VP node through an edge labeled MO Since we do not know anything about the modifying node whether it is a non terminal node or a token we simply use the node element as a place holder This element can match any node or annotation in the graph cat VP amp node amp 1 gt tiger func MO 2 11 It is also possible to negate the label of the dominance edge as in the following query cat VP amp node amp 1 gt tiger func MO 2 which finds all VPs dominating a node with a label other than MO 4 6 Searching for Pointing Relations Coreference and Dependencies Pointing relations are used to express an arbitrary directed relationship between two elements terminals or non terminals without implying dominance or coverage inheritance For instance in the pcc2 demo corpus elements in the mmax namespace may point t
3. de hub corpling Pepper pepperParams gt lt PepperJobParams id 1 gt lt importerParams formatName PAULA formatVersion 1 0 sourcePath file PEPPER_HOME examples samplel paula pcc2 gt lt exporterParams formatName relANNIS ROMANE Vie SEO laws mon destinationPath file PEPPER_HOME examples samplel relANNIS gt lt PepperJobParams gt lt PepperParams PepperParams gt 23
4. eeeeeeeeeeeseeesseeeeeeeees 17 5 1 Triggering VS A ace ees cece actg tered csaus asec ideas ays date Gute dacertddee andduaneey 17 5 2 Visualizations with Software Requirements eeseseeeseeeeeseeseesressreeresressereresresse 18 6 Converting Corpora for ANNIS using Pepper 1 0 0 0 eecceeeeeceeeeeeeeeeeeeeeeeeeeeeeeees 20 Gd Tips ii PRD POE cs esiotsas enedes E oE S EEE EE Eni EE Ei 20 6 2 R nning Pepp f sssisssiecrieieinsi iiini aniisi esisin Tesia ieee 20 6 3 Pepper Workflow ssccoxcstunceaccianscanspandaniaaiceastandeaeinssniatanearoniena teatieaeeneeRaeeee 20 OFF MUAY coo tesa eas E E E E eae 22 1 Introduction ANNIS2 is an open source browser based search and visualization architecture for multi layer corpora It can be used to search for complex graph structures of annotated nodes and edges forming a variety of linguistic structures such as constituent or dependency syntax trees coreference and parallel alignment edges span annotations and associated multi modal data audio video This guide provides an overview of the current ANNIS2 system first steps for installing either a local instance or an ANNIS server with a demo corpus as well as tutorials for converting data for ANNIS and running queries with AQL ANNIS Query Language 2 New Features in Version 2 1 8 Performance Massive acceleration of query times through a completely restructured DB scheme Faster corpus import times by using more efficient indexes Imp
5. path to the file or path they are supposed to import from or export to Modeling a Workflow via XML An xml file defining a module is called a Pepper workflow file and has the ending pepperparams A workflow description using module names for identification looks as follows lt xml version 1 0 encoding UTEF 8 gt lt PepperParams PepperParams xmi version 2 0 xmlins xmi http www omg org XMI xmins PepperParams de hub corpling Pepper pepperParams gt lt PepperJobParams id 1 gt lt importerParams moduleName _sourcePath specialParams gt SS ae lt moduleParams moduleName specialParams gt Mitt ee a i eee lt exporterParams moduleName destinationPath specialParams gt SSS a nE lt PepperJobParams gt LSS 54 SSR lt PepperParams PepperParams gt The xml element PepperJobParams stands for a Pepper job One job does one conversion you can specify one or more jobs in one workflow file Every job has to have a unique id and has to contain at least one importer description and one exporter description A manipulator description is optional There is no upper limit for the number of module descriptions which can be used for a conversion The attribute moduleName identifies the module which is to be used for the current step Importers have an attribute sourcePath where you have to specify the path of the source corpus Exporters have an attribute destinat
6. 8 gt lt PepperParams PepperParams xmi version 2 0 xm Ins xmi http www omg org XMI xm ins PepperParams de hub corpling Pepper pepperParams gt lt PepperJobParams id 1 gt 22 lt importerParams moduleName PAULAImporter sourcePath file PEPPER_HOME examples samplel paula pcc2 _ gt lt exporterParams moduleName RelANNISExporter destinationPath file PEPPER_HOME examples samplel relANNIS gt T lt PepperJobParams gt lt PepperParams PepperParams gt This file also can be found under PEPPER_HOME examples sample 1 paula2relANNIS pepperParams For testing you can call pepperStart bat p PEPPER_HOME examples samplel paula2relANNIS pepperParams or bash pepperStart sh p PEPPER_HOME examples samplel paula2relANNIS pepperParams Take care to replace PEPPER HOME with the absolute path of the pepper directory After doing this you will find the newly created folder relANNIS in PEPPER_HOME examples samplel relANNIS which contains the pcc2 corpus in the relANNIS format The following example will show a similar workflow producing exactly the same result but here instead of identifying the PepperModule by using the name we use the format name and the format version lt xml version 1 0 encoding UTF 8 gt lt PepperParams PepperParams xmi version 2 0 xmilns xmi http www omg org XMI xmins PepperParams
7. ANNIS2 User Guide Version 2 1 8 For the latest documentation see also http korpling german hu berlin de trac Contents Ca 078 Liye 105 1 resent mrtnent een pe nnn a PRON rr nett E ere ye nt eae omer pr rt Carrere ne yee 1 2 New Features in Version 220 sens soncecascecceticcteidoasasadaniaaroiednoces lt ephdodaecsuagnatentesudebanteacells 1 3 Installing ANNIS2 ga epica se cence ten cenae ee idence eta ta ns eeo aa a o aS Ee 2 3 1 Installing a Local Version ANNIS Kickstarter 0 eecceescceeeeeceeeseeeeeeeeeeseeeenes 2 3 2 Building and Installing an ANNIS Server eee eeeesseeseeeeeeceaeceeeeseeeeneeenaeenes 3 4 Ronnie Queries mi ANNIS aireicirinertenoraeioririi e aA A 5 4 1 Th ANNIS2 Titer laces svcssisiedatageanensasnevsaaasatien EERTE AE OERA 5 4 2 Using the ANNIS2 Query Builder seeeseseseeeeseseeseeseesrrsrersersesressressrseresresseseres 7 43 Searching for Word ForS issereosac reses er ier E EE a E IEE 8 4 4 Searching for Annotations scscassersannscracstentacspanreatdenneniaeneadiartiensea teateaesncneeatieees 9 4 5 SCAM MMI tor Tr S ecis ekna aeiiae esa iena aiai 10 4 6 Searching for Pointing Relations Coreference and Dependencies 12 4 7 Exporting Search ROS U MS sip acecctnct scasccchaccesecencensgevndeiesa ates vadboneemersceansecticeangienee 13 BS Complete List ot Oper nis csixigseoittiga se aaccgsesaievssceck co a a E Geen 15 5 Configuring Visualizations with the Resolver Table ieee
8. art bat p workflow file e Unix Linux MacOS bash pepperStart sh p workflow file The content of the workflow file is described in the following section 6 3 Pepper Workflow The worklfow of a conversion process in Pepper consists of three phases An import phase a manipulation phase and an export phase e In the import phase modules called importers map data from an input format to Salt the metamodel used to describe all types of data e Inthe manipulation phase modules called manipulators map data from one Salt model to another Salt model to alter data e g by renaming certain annotation names e In the export phase modules called exporters map data from Salt to an export format Each phase can include several steps The export phase and the import phase can include 1 to n steps whereas the manipulation phase can include 0 to n steps Steps are the lifecycles of running a module i e a PepperModule Every module can be identified by a name the module name In addition importers and exporters also can be identified by a pair consisting of the format name and the format version they support During 20 processing Pepper searches for a module with a given module name or a given pair of format name and format version and starts it Additionaly for every module you can add a file with parameters for this module Please see the description of the module you want to use for details Importers as well as exporters also needs a
9. e oe available immediately after login In the middle the list of currently available corpora is shown Using the quer Buide sow checkboxes on the left of each corpus it is possible to rest Vaid Query 4 select which corpora should be searched in hold down a More Corpora shift to select multiple corpora simultaneously If you ane Se OE cannot see a corpus that should be available to you or cemanvisctronicrreey a osa else if the corpora list is too cluttered you may click on ees 29 e arabic test i more corpora to open the corpora window You may H si lt asbani sudo 3 n then drag and drop the desired or unwanted corpora 22ssneesoeu 2 wi 3 i T pees 3 s3 G between the list and the window ala ea Search J BA l The AnnisQL field at the top of the form is used for Context Left 5 v inputting queries manually see the tutorials on the contexerisnt 5 ANNIS Query Language As soon as a one or several Resutts per page 10 Show Result corpora are selected and a query is entered or modified the query will be validated automatically and possible errors in the query syntax will be commented on in the Result box below When modifying a query a delay of two seconds ia activated before the query is re sent to the server for validation Once a valid query has been entered pressing the Show Result button will retrieve the number of matching positions in the sel
10. e span of tokens covered by each 13 annotation may optionally be given in square brackets to turn this off use the optional parameter numbers false The user can specify annotation layers to be exported in the additional Parameters box using the setting keys and annotation names separated by comas If nothing is specified in the parameters box all available annotations will be exported Multiple options are separated by a semicolon e g keys tok pos cat numbers false An example output with token numbers looks as follows 1 tok ein Dialog zwischen den Generationen angesto en cat NP 1 5 S 1 6 VP 1 6 PP 3 5 pos ART 1 1 NN 2 2 APPR 3 3 ART 4 4 NN 5 5 VVPP 6 6 7 7 Meaning that the annotation cat NP applied to tokens 1 5 in the search result and so on Note that when specifying annotation layers if the reserved name tok is not specified the tokens themselves will not be exported annotations only The WekaExporter outputs the format used by the WEKA machine learning tool http www cs waikato ac nz ml weka Only the attributes of the search elements 1 2 etc in AQL are outputted and are separated by commas The order and name of the attributes is declared in the beginning of the export text as in this example relation name attribute 1l_id string attribute 1_token string attribute 1_tiger cat string attribute 2_id string attribute 2_token string attribute 2_tiger lemma
11. ected corpora in the Result box and open the Result Window to display the first set of matches The context surrounding the matching expressions in the result list ist determined by the context left and context right options at the bottom of the search form and can be set to up to 10 tokens on each side though some corpora allow longer spans such as entire texts to be viewed using special discourse visualizations The Result Window The result window shows search results in pages of 10 hits each by default this can be changed in the Search Form The toolbar at the top of the window allows you to navigate between these pages The Token Annotations button on the toolbar allows you to toggle the token based annotations such as lemmas and parts of speech on or off for you convenience The Citation URL button provides a hyperlink which you can e mail or cite allowing others to reproduce your query der wie eine Mumie auf der Bank der vie ein Mumie aut der Bank ART KOKOM ART NN APPR ART NN n Sq Mast Nom Sg Fem Nom Sg Fem Dat SqFem Dat Sg Fem tiger morph Nom Sg Fem The result list itself initially shows a KWIC key word in context concordance of matching positions in the selected corpora with the matching region marked red and the context in black on either side Token annotations are displayed in gray under each token and hovering over them with the mouse will show the annotation name and namespace M
12. he middle allows you to choose between an exact match the symbol or wildcard search using Regular Expressions the symbol The annotation value is given on the right and should NOT be surrounded by quotations see the example below It is also possible to specify multiple annotations applying to the same position by clicking on Add multiple times Clicking on Clear will delete the values in the node To search for word forms simply leave the field name on the left empty and type directly on the right under Value A node with no data entered will match any node that is an underspecified token or non terminal node or annotation To specify the relationship between nodes first click on the Edge Add Clear x Edge button at the top left of one node and then click the Dock button which becomes available on the other nodes An edge will connect the nodes with an extra box from which operators may be selected see below For operators allowing additional labels e g the dominance operator gt posi ion refeenti roles Field op Yalue allows edge labels to be specified you may type directly into the edge s operator box as in the example with a func label in the image below Note that the node clicked on first where the Edge button was clicked will be the first node in the resulting quey i e if this is the first node it will dominate the second node 1 gt 2 and not the other way around as also represe
13. indirectly by wie A range of allowed distances can also be specified numerically as follows Ss tatisch amp wie amp 1 1 5 2 Meaning the two words may appear at a distance of 1 to 5 tokens The operator allows a distance of up to 50 tokens by default so searching with 1 50 is the same as using instead Greater distances e g 1 100 for within 100 tokens should always be specified explicitly Finally we can add metadata restrictions to the query which filter out documents not matching our definitions Metadata attributes must be preceded by the prefix meta and may not be bound i e they are not referred to as 1 etc and the numbering of other elements ignores their existence Ss tatisch amp wie amp 1 1 5 2 amp meta Genre Sport To view metadata for a search result or for a corpus press the i icon next to it in the result window or in the search form respectively 4 4 Searching for Annotations Annotations may be searched for using an annotation name and value The names of the annotations vary from corpus to corpus though many corpora contain part of speech and lemma annotations with the names pos and lemma respectively annotation names are case sensitive For example to search for all forms of the German verb sein to be in a corpus with lemma annotation such as PCC2 simply select the PCC2 corpus and enter lemma sein Negative searches are also possible using instead of
14. ionPath where you have to specify the path of the destination corpus The attribute specialParams can be used for parameters for the current module SpecialParameters must be given in a property file 21 Caution Please make sure that every path is in URI syntax and is an absolute path A workflow description using format name and format version for identification of im and exporters looks as follows lt xml version 1 0 encoding UTF 8 gt lt PepperParams PepperParams xmi version 2 0 xmins xmi http www omg org XMI xmins PepperParams de hub corpling Pepper pepperParams gt lt PepperJobParams id 1 gt lt importerParams formatName formatVersion sourcePath specialParams gt CUS p49 Se lt moduleParams moduleName specialParams gt lt Ies pag Se lt exporterParams formatName formatVersion sourcePath _destinationPath specialParams gt Pp SHS goo ee lt PepperJobParams gt SIS Soy Se lt PepperParams PepperParams gt Unlike the upper example here we use the attributes formatName and formatVersion to ident ify an importer as well as an exporter 6 4 Example In PEP PER_HOME you will find a folder examples with a small sample corpus for conversion this is the pcc2 demo corpus in the PAULA XML format The following workflow file defines the conversion of this corpus from PAULA to the relANNIS format lt xml version 1 0 encoding UTF
15. ips between potentially overlapping annotations in spans For example the operator _i_ examines whether one annotation fully contains the span of another annotation the i stands for includes Topic ab amp Inf Stat new amp 1 _i 2 This query finds aboutness topics Topic ab containing information structurally new discourse referents 4 5 Searching for Trees In corpora containing hierarchical structures annotations such as syntax trees can be searched for by defining terminal or none terminal node annotations and their values A simple search for prepostional phrases in the small PCC2 demo corpus looks like this 10 tiger cat PP If the corpus contains no more than one annotation called cat the optional namespace in this case tiger may be dropped This finds all PP nodes in the corpus To find all PP nodes directly dominating a proper name a second element can be specified with the appropriate part of speech pos value cat PP amp pos NE amp 1 gt 2 The operator gt signifies direct dominance which must hold between the first and the second element Once the Result Window is shown you may open the tiger annotation level to see the corresponding tree i Mit seinem Tor zum 1 0 f r die Ukraine mit sein Tor zu 1 0 f r der Ukraine APPR PPOSAT NN APPRART CARD APPR ART NE Dat Sg Neut Dat Sq Neut Dat Sg Neut Acce sgFem Acc sgFe S tiger gt md AC NK NK MNR AC Mk MNR o ad
16. lAnnis Select this directory but do not go into it and press OK Once import is complete press Launch Annis frontend and login with the username and password test to test the corpus try selecting the pcc2 corpus typing pos NN in the AnnisQL box and clicking Show Result See the section Running Queries in ANNIS2 in this guide for some more example queries or press the Tutorial button at the top left of the interface 3 2 Building and Installing an ANNIS Server The ANNIS server version can be installed on UNIX based server or else under Windows using Cygwin the freely available UNIX emulator To install the ANNIS server l Install a PostgreSQL server for your operating system from http www postgresql org download 2 Install a web server such as Tomcat or Jetty 3 Make sure you have JDK 6 and Maven 2 or install them if you don t 4 If you re using Cygwin and Windows you will also need to install the patch program via the Cygwin package manager 5 Download and unzip Annis 2 1 8 zip then run the following commands replacing the appropriate directories cd lt unzipped source Annis Service mvn DskipTests true install mvn DskipTests true assembly assembly tar xzvf target annis service lt version gt distribution tar gz C lt installation directory gt 6 Next initialize your ANNIS database only the first time you use the system 7 Set the environment variables each time
17. lation A B gt LABEL n m for relation chains of length nto m 15 A left most ce child Bxy A right most i child xyB Common X parent node AB x x Specifies the amount of directly x arity n rity LIN dominated children that the searched 1 n node has x Specifies the length of the span of okens covered by the node x length n Length node x is the root of a subgraph i e it is not dominated by any node 16 5 Configuring Visualizations with the Resolver Table 5 1 Triggering Visualizations By default ANNIS2 displays all search results in the Key Word in Context KWIC view in the search result window Further visualizations such as syntax trees or grid views are displayed by default based on the following namespaces Nodes with the namespace tiger tree visualizer Nodes with the namespace exmaralda grid visualizer Edges with the namespace mmax discourse view Nodes with the namespace external multimedia player In these cases the namespaces are usually taken from the source format in which the corpus was generated and carried over into relAnnis during the conversion It is also possible to use other namespaces most easily when working with PAULA XML In PAULA XML the namespace is determined by the string prefix before the first period in the file name paula_id of each annotation layer In order to manually determine the visualizer and the display name for each na
18. lid null values apply to all corpora 17 namespace specifies relevant namespace which triggers the visualization element determines if a node or an edge should carry the relevant annotation for triggering the visualization vis_type determines the visualizer module used and is one of tree constituent syntax tree grid annotation grid with annotations spanning multiple tokens old_grid deprecated version of grid discourse a view of the entire text of a document possibly with interactive coreference links arch_dependency dependency tree with labeled arches between tokens requires SVG enabled browser see 5 2 ordered_dependency arrow based dependency visualization for corpora with dependencies between non terminal nodes requires GraphViz see 5 2 hierarchical_dependency unordered vertical tree of dependent tokens requires GraphViz graph a debug view of the annotation graph requires GraphViz see 5 2 file a linked multimedia file The additional system internal debug views paula and paula_text deliver an XML representation of hits and entire texts respectively display_name determines the heading that is shown for each visualizer in the interface order determines the order in which visualizers are rendered in the interface low to high mappings provides additional parameters for some visualizations tree the annotation names to be displayed i
19. lt set is very large 14 4 8 Complete List of Operators The ANNIS Query Language AQL currently includes the following operators Operator Description IIlustration Notes For non terminal nodes precedence is determined by the right most and left most terminal children For specific sizes of precedence spans n m Can be used e g 3 4 between 3 and 4 token distance A specific edge type may be specifed A e g gt secedge to find secondary direct edges Edges labels are specified in brackets e g gt func 0A for an edge with the function object accusative For specific distance of dominance gt n m can be used e g gt 3 4 dominates with 3 to 4 edges distance B identical Applies when two annotation cover the coverage exact same span of tokens AAA Applies when one annotation covers a i inclusion B span identical to or larger than another overlap AAA For overlap only on the left or right i BBB side use _ol_ and _or_ respectively AAA Both elements span an area beginning 1 ere angneg BB with the same token Both elements span an area ending night angned BBB with the same token A labeled directed relationship between two elements Annotations can be specified with gt LABEL annotation VALUE gt TARET pz An indirect labeled relationship indirect between two elements The length of gt LABEL pointing the chain may be specified with re
20. mespace in each corpus the resolver table in the database must be edited To do so open PGAdmin or if you did not install PGAdmin with ANNIS then via PSQL and access the table resolver_vis_map it can be found in PGAdmin under PostgreSQL 8 4 gt Databases gt anniskickstart gt Schemas gt public gt Tables for ANNIS servers replace anniskickstart with annis_db You may need to give your PostgreSQL password to gain access Right click on the table and select View Data gt View All Rows The table should look like this E Edit Data PostgreSQL 8 4 localhost 5432 anniskickstart resolver_ is_map File Edit View Tools Help corpus version namespace element vis_type display_name order mappings character character yar character var character var character var character var numeric character var tiger tree tree exmaralda grid exmaralda mmax grid mmax 103 mmax discourse coref 104 urml old_grid urml 105 external File external file paula paula 107 paula_text paula text b3 parses 1 bitpar tree bitpar 1 b3 parses 1 lingenio tree lingenio 2 parallel_tree_ tiger de tree Syntax Germar 1 parallel_tree_ tiger en tree Syntax English 2 b3 parses 1 discourse Whole Text 4 SMULTRON_E german tree Syntax Germar 1 SMULTRON_E english tree Syntax English 2 Resolver table resolver_vis_map The columns in the table can be filled out as follows corpus determines the corpora for which the instruction is va
21. n non terminal nodes can be set e g using node_key cat for an annotation called cat the default and similarly the edge labels using edge_key func for an edge label called func the default Instructions are separated using semicolons graph use ns_all true to visualize the entire annotation graph Specifying e g node_ns tiger or edge_ns tiger instead causes only nodes and edges of the namespace tiger to be visualized i e only a subgraph of all annotations the field version is reserved for future development 5 2 Visualizations with Software Requirements Some ANNIS visualizers require additional software depending on whether or not they render graphics as an image directly in Java or not At present three visualizations require an _ installation of the freely available software GraphViz http www graphviz org ordered_dependency hierarchical_dependency and the general graph visualization To use these install GraphViz on the server or your local machine for Kickstarter and make sure it is available in your system path check this by calling e g the program dot on the command line Another type of restriction is that some visualizers may use SVG scalable vector graphics instead of images which mean the user s browser must be SVG capable e g 18 FireFox or a plugin must be used e g for Internet Explorer 8 or below This is currently the case for the arch_dependency visualizer amp hierarchical de
22. nted by the arrows along the edge Add Clear X Field op Yalue r amp cat 5 gt tiger func SB v pS Field op Yalue cat pp x Add Clear X r 4 3 Searching for Word Forms To search for word forms in ANNIS2 simply select a corpus in this example the small PCC2 demo corpus and enter a search string between double quotation marks e g statisch Note that the search is case sensitive so it will not find cases of capitalized Statisch for example at the beginning of a sentence In order to find both options you can either look for one form OR the other using the pipe sign statisch Statisch or else you can use regular expressions which must be surrounded by slashes instead of quotation marks Ss tatisch To look for a sequence of multiple word forms enter your search terms separated by amp and then specify that the relation between the elements is one of precedence as signified by the period operator so amp statisch amp 1 2 The expression 1 2 signifies that the first element so precedes the second element statisch For indirect precedence where other tokens may stand between the search terms use the operator Ss o amp statisch amp wie amp 1 2 amp 2 3 The above query finds sequences beginning with either So or so followed directly by statisch which must be followed either directly or
23. o each other to express coreference or anaphoric relations The following query searches for two np_form annotations which specify for example whether a nominal phrase is pronominal definite or indefinite mmax np_form pper amp mmax np_form defnp amp 1 gt anaphor_antecedent 2 Using the pointing relation operator gt with the type anaphor_antecedent the first np_form which should be a personal pronoun pper is said to be the anaphor to its antecedent the second np_form which is definite defnp To see a visualization of the coreference relations open the mmax annotation level in the example corpus In the image below one of the matches for the above query is highlighted in red die Seeburger und einige Grof Glienicker sie the Seeburgers and some Grof Glienickers they Other discourse referents in the text marked with an underline may be clicked on causing coreferential chains containing them to be highlighted as well Note that discourse referents may overlap leading to multiple underlines Die Seeburger the Seeburgers is a shorter discourse referent overlapping with the larger one the Seeburgers and some Grof Glienickers and each referent has its own underline Annotations of the coreference edges of each relation can be viewed by hovering of the appropriate underline mmax discourse Steilpass Wunder gibt es immer wieder Erst spielen die Daigone Gemeindevertreter so statisch und
24. om http www postgresql org download and make a note of the administrator password you set during the installation After installation Postgres may automatically launch the Postgres Stack Builder to download additional components you can safely skip this step and cancel the Stack Builder if you wish You may need to restart your OS if the Postgres installer tells you to Download and unzip Annis Kickstarter 2 1 8 zip from the ANNIS website Start AnnisKickstarter bat if you re using Windows or run the bash script AnnisKickstarter sh otherwise this may take a few seconds the first time you run Kickstarter At this point your Firewall may try to block Kickstarter and offer you to unblock it do so and Kickstarter should start up m Note for most users it is a good idea to give Java more memory if this is not already the default You can do this by editing the script AnnisKickstarter and typing the following after the call to start java before splash splashscreen gif Xss1024k Xmx1024m To accelerate searches it is also possible to give the Postgres database more memory see the link in the next section below Once the program has started if this is the first time you run Kickstarter press Init Database and supply your PostGres administrator password from step 1 Download and unzip the pcc2 demo corpus from the ANNIS website Press Import Corpus and navigate to the directory containing the directory pec2_re
25. ore complex annotation levels can be expanded if available by clicking on the plus icon next to the level s name e g tiger and exmaralda for the annotations in the oG cR oA D S lt lt lt lt d NK NK NK NP NK Komik Erwachsene Karola Andr Select Displayed Annotation Levels 7 D int stat om caen tree and grid views in the picture to ne me e NP the right circled in red PP Sent s 4 2 Using the ANNIS2 Query Builder To open the graphical query builder click on the Query Builder Show gt gt button on the Search Form then clicking Query Builder hide lt lt will close the Query Builder On the left hand side of the toolbar at the top of the query builder canvans you will see the Create Node button Use this button to define nodes to be searched for tokens non terminal nodes or annotations Creating nodes and modifying them on the canvas will immediately update the AnnisQL field in the Search Form with your query though updating the query on the Search Form will not create a new graph in the Query Builder ANNIS Tutorial Search Form AnnisQL Create Node Query Builder Result mpty Query More Corpora Name Texts Tokens Aischylos Persai 1 6212 D PCC176_v1 3 113469 12296 i In each node you create you may click on Add to specify an annotation value The annotation name can be typed in or selected from a drop down list The Op erator field in t
26. ort speed is no longer dependent on current DB size previous corpora thanks to partitioning Visualization Three new dependency tree visualizations o ordered arch dependency visualizer courtesy of Kim Gerdes ILPGA Paris o unordered and ordered tree dependency visualizers developed in conjunction with Dag Haug PROIEL project Oslo Debug graph visualization a graph of all annotations not very readable but useful for debugging Improved discourse coreference visualizer now handles nested discourse referents and displays edge annotations Improved grid visualizer handles overlapping spans with same annotation name correctly Differential match coloring each search node is highlighted in a different color Search and export Addition of generations for pointing relations 1 gt anaphor 3 4 2 finds chains of length 3 to 4 New exporters for tokens with their annotations and flattened grids in plain text Negation in metadata works correctly bugfix Backend timeout works correctly bugfix Simplification of AQL SQL generation mainly relevant for developpers For change logs of previous version see their respective distributions or user guides 3 Installing ANNIS2 3 1 Installing a Local Version ANNIS Kickstarter Local users who do not wish to make their corpora available online can install ANNIS Kickstarter To install Kickstarter follow these steps 1 Download and install PostgreSQL 8 4 for your operating system fr
27. pendency Wunder gibt es immer wieder Hierarchical dependency visualizer Arch dependency visualizer SJ ordered dependency Kat i6wv Inoovs Thy niotiw a t v A yer T NAPA VTIK Ordered tree dependency visualizer 19 6 Converting Corpora for ANNIS using Pepper 1 0 ANNIS2 uses a relational database format called relANNIS The Pepper converter framework allows users to convert data from PAULA XML EXMARaLDA XML Tiger XML and TreeTagger directly into relAnnis the Tiger XML conversion is limited to corpora without secondary edges at the moment Further formats including Tiger XML with secondary edges can be converted first into PAULA XML and then into relANNIS using the converters found on the ANNIS downloads page 6 1 Installing Pepper Unzip the file Pepper_1 0 0 zip Pepper is now ready to run If this does not work correctly you can compile the sources by running an ANT script for which you will need to install ANT With ANT installed change the directory to your PEPPER_HOME and run ant f build xml 6 2 Running Pepper To run Pepper you have to create a workflow containing the steps to be carried out during the conversion process The workflow should be described in an xml file called Pepper workflow or Pepper params To run the program you must assign the workflow file by using the flag p in program call The following example shows the usage e Windows pepperSt
28. rch visualizer which shows the verb gibt gives and its object Wunder miracles Wunder gibt es immer wieder 4 7 Exporting Search Results By going to the Export tab at the bottom of the search form on the left you can select one of several exporters Search Export Exporter GridExporter ad Context Left 5 MA Context Right 5 hl Parameters keys tok cat pos Perform Export The SimpleTextExporter simply gives the text for all tokens in each search result including context in a one row per hit format The tokens covered by the match area are marked with square brackets and the results are numbered as in the following example 1 Tor zum 1 0 fiir die Ukraine stiirzte der 1 62 Meter groBe 2 der 1 62 Meter groBe Gennadi Subow die deutsche Nationalelf vor bergehend in 3 und Reputation kampfenden Mannschaft von Rudi V ller der Weg zur Weltmeisterschaft 4 Reputation kampfenden Mannschaft von Rudi V ller der Weg zur Weltmeisterschaft endgiiltig 5 die deutschen Nationalkicker einen Rudi Riese auf der Bank The TextExporter adds all annotations of each token separated by slashes e g dogs NN dog for the token dogs annotated with a part of speech NN and a lemma dog The GridExporter adds all annotations available for the span of retrieved tokens with each annotation layer in a separate line Annotations are separated by spaces and the hierarchical order of annotations is lost though th
29. string attribute 2_tiger morph string attribute 2_tiger pos string data 288662 NU 289175 NU 289660 NU 288672 NU 289614 NU 289625 NU 288607 NU 288620 NU 289220 NU 288610 NU 289174 NU 289611 NU 288624 NU NP 288392 ganze ganz Pos Acc Sg Fem ADJA NP 288712 geladenen geladen Pos Nom P1 ADJA L NP 289409 D beritzer D beritzer Pos ADJA NP 288302 deutschen deutsch Pos Nom P1 Masc ADJA NP 289291 deutsche deutsch Pos Nom Sg Fem ADJA L NP 289245 fulminanter fulminant Pos Nom Sg Masc ADJA NP 288242 einstige einstig Pos Nom Sg Fem ADJA L NP 288334 hnliche hnlich Pos Acc Pl Neut ADJA L NP 288883 groBe groB Pos Nom Sg Fem ADJA L NP 288313 deutsche deutsch Pos Acc Sg Fem ADJA L NP 288809 bd se bd se Pos Nom Sg Fem ADJA L NP 289241 Dallgower Dallgower Pos ADJA L NP 288330 ukrainische ukrainisch Pos Nom Sg Masc ADJA TE ee T TE a ae E T 3 E The export shows the properties of an NP node dominating a token with the part of speech ADJA Since the token also has other attributes such as the lemma the token text and morphology these are also retrieved Note that exporting may be slow in both exporters if the resu
30. utorial AnnisQL fok amp tok amp 1 gt dep Page 1 ots gt DL amp Token Annotations Show Citation URL Displaying Results 1 10 of 43 func 0A 2 amp cat S amp 3 _i_ 1 amp node amp 3 al gt secedge 4 P correction correcting oot Query Builder Show gt gt Result 43 w hrend 78 Prozent sich fur ahrend 78 P sich f r PRF APPR Acc PI More Corpora J dependencies Name Texts Tokens FalkoEssayL2V2_0 248 131511 ONTONOTES_v1 5_small 4 6450 SMULTRON _ Banana 2 3782 E TueBaS_no_cye 2187 770949 agni 24 184 b4tatian2 0 2031 w hrend 78 Prozent sich f r Bush und vier Prozent f r Clinton aussprachen pec 3 3 S constituents pcc2 2 tiger1 dep 1 929 tiger2 1971 888578 NK w hrend 78 c j vier Prozent f r Die Vase auf dem Tisch ist gr er als die Vase f Search Export tal 3 animacy grid Context Left 0 Select Displayed Annotation Levels Context Right a mmax ref_type inanim mmax ref_type Results per page 10 tok ist i B coreference discourse Show Result Deasa dem Tisch ist gr er als die Vase auf der Fensterbank Ich finde Bi sieht nicht so gut aus weil der Tisch zu klein ist The ANNIS2 interface is comprised of several windows the most important of which are the search form in the red box above and the results window in the blue box above The Search Form ZED The Search Form on the left of the interface window is ea
31. verzagt wie m die deutsche Abwehmeihe der Fubballkicker Und dann kommt aus der Tiefe solch ein fulminanter Steilpass von dem man hofft dass Mie Seebumgen oder Grob Glienicker RAPSpieler ihn aufnehmen k nnen En Befreiungsschlag ist es allerdings nicht weil es vorerst keine Gefahr f rs DallgoW r Tor gab Die Seeburger und einige Grok Glienicker haben den Ball erst zunickgespielt und dann um so drangender wieder gefordert Nun sollen Bie zeigen wie iig die Chance verwerten Gne Diskussion wo k nftig die Trainerkabine stehen soll ware in der jetzigen Spielsituation verheerend Und eine Parallele zu den deutschen Grotten Kickem gibt es immer noch Auch wenn die Spieler aus den verschiedenen ereinen zusammengewtirfelt sind Big m ssen sich daran gew hnen dass Big nun in einer Mannschaft D bertzer Heide spielen Und das heit gemeinsam und nicht gegeneinander Ermahnungen won der w Another way to use pointing relations is found in syntactic dependency trees The queries in this case can use both pointing relation types and annotation as in the following query pos VVFIN amp tok amp 1 gt dep func obja 2 12 This query searches for a finite verb with the part of speech VVFIN and a token with a pointing relation of the type dep for dependency between the two annotated with func obja the function Object Accusative The result can be viewed with the dependency a
32. when starting up export ANNIS_HOME lt installation directory gt export PATH SPATH SANNIS_HOME bin 8 Now you can import some corpora annis admin sh import path to corpusl path to corpus2 Important The above import command calls other PostgreSQL database commands If you abort the import script with Ctrl C these SQL processes will not be automatically terminated instead they might keep hanging and prevent access to the database The same might happen if you close your shell before the import script terminates so you will want to prefix it with the nohup command 9 Now you can start the ANNIS service annis service sh start 10 To get the Annis front end running first compile it cd lt unzipped source gt mvn DskipTests true install If no error occurs the war file will be available under lt unzipped source gt Annis web target Annis web war 11 And configure your web server as described here http korpling german hu berlin de trac annis wiki Documentation Web Tomcat The latest instructions for compiling and installing the ANNIS Server can also be found at http korpling german hu berlin de trac annis wiki Documentation We also strongly recommend reconfiguring the Postgres server s default settings as described here http korpling german hu berlin de trac annis wiki Documentation Service PostgreSQL 4 Running Queries in ANNIS2 4 1 The ANNIS2 Interface ANNIS T
Download Pdf Manuals
Related Search