{ "paper_id": "I13-1007", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T07:15:19.933487Z" }, "title": "(Pre-)Annotation of Topic-Focus Articulation in Prague Czech-English Dependency Treebank", "authors": [ { "first": "Ji\u0159\u00ed", "middle": [], "last": "M\u00edrovsk\u00fd", "suffix": "", "affiliation": { "laboratory": "", "institution": "Charles University", "location": { "settlement": "Prague" } }, "email": "" }, { "first": "Kate\u0159ina", "middle": [], "last": "Rysov\u00e1", "suffix": "", "affiliation": { "laboratory": "", "institution": "Charles University", "location": { "settlement": "Prague" } }, "email": "" }, { "first": "Magdal\u00e9na", "middle": [], "last": "Rysov\u00e1", "suffix": "", "affiliation": { "laboratory": "", "institution": "Charles University", "location": { "settlement": "Prague" } }, "email": "" }, { "first": "Eva", "middle": [], "last": "Haji\u010dov\u00e1", "suffix": "", "affiliation": { "laboratory": "", "institution": "Charles University", "location": { "settlement": "Prague" } }, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "The objective of the present contribution is to give a survey of the annotation of information structure in the Czech part of the Prague Czech-English Dependency Treebank. We report on this first step in the process of building a parallel annotation of information structure in this corpus, and elaborate on the automatic pre-annotation procedure for the Czech part. The results of the pre-annotation are evaluated, based on the comparison of the automatic and manual annotation.", "pdf_parse": { "paper_id": "I13-1007", "_pdf_hash": "", "abstract": [ { "text": "The objective of the present contribution is to give a survey of the annotation of information structure in the Czech part of the Prague Czech-English Dependency Treebank. We report on this first step in the process of building a parallel annotation of information structure in this corpus, and elaborate on the automatic pre-annotation procedure for the Czech part. The results of the pre-annotation are evaluated, based on the comparison of the automatic and manual annotation.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "In the past three or four decades, topic-focus articulation (known also as sentence information structure) is a language phenomenon that has attracted an enormous interest in linguistics and has become a \"hot\" topic of linguistic studies. No wonder then, that these days several linguistic teams (e.g. at the University of Potsdam, University of Berlin, University of Stuttgart, Charles University in Prague) have attempted to include the annotation of information structure in the annotating schemes they propose. Among corpora that contain also annotation of information structure or such type of annotation is planned in them there are e.g. ANNIS database (Annotation of Information Structure, see Dipper et al., 2004) , The English Switchboard Corpus (see Calhoun et al., 2005) , the corpus DannPASS (Danish Phonetically Annotated Spontaneous Speech, see Paggio, 2006) and the Prague Dependency Treebank (for the information on PDT, see Haji\u010d et al., 2006) .", "cite_spans": [ { "start": 701, "end": 721, "text": "Dipper et al., 2004)", "ref_id": "BIBREF4" }, { "start": 760, "end": 781, "text": "Calhoun et al., 2005)", "ref_id": "BIBREF1" }, { "start": 859, "end": 872, "text": "Paggio, 2006)", "ref_id": "BIBREF14" }, { "start": 941, "end": 960, "text": "Haji\u010d et al., 2006)", "ref_id": "BIBREF6" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "There are also several types of annotation guidelines and schemes for the different corpora, based on various linguistic theories dealing with information structure (e.g. Haji\u010dov\u00e1 et al., 2000; Nissim et al., 2004; Dipper et al., 2007; Donhauser, 2007; Cook and Bildhauer, 2011) .", "cite_spans": [ { "start": 171, "end": 193, "text": "Haji\u010dov\u00e1 et al., 2000;", "ref_id": "BIBREF9" }, { "start": 194, "end": 214, "text": "Nissim et al., 2004;", "ref_id": "BIBREF13" }, { "start": 215, "end": 235, "text": "Dipper et al., 2007;", "ref_id": null }, { "start": 236, "end": 252, "text": "Donhauser, 2007;", "ref_id": "BIBREF5" }, { "start": 253, "end": 278, "text": "Cook and Bildhauer, 2011)", "ref_id": "BIBREF2" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "In our paper, we present the annotation of topic-focus articulation in the Czech part of the Prague Czech-English Dependency Treebank, based on the theory of topic-focus articulation as developed withing the Praguian Functional Generative Description. It is the first step in the process of building a parallel Czech-English corpus annotated with this type of linguistic information. 1", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "The first complex and consistent theoretically-based annotation of topic-focus articulation was already fully applied in the first Czech corpus from the Prague corpora family, the Prague Dependency Treebank (PDT; Haji\u010d et al., 2006 , updated in Bej\u010dek et al., 2012 , and is available for the linguistic community. PDT is a large collection of Czech journalistic texts, (basically) manually annotated on several layers of language description (more than 3 thousand documents consisting of almost 50 thousand sentences are annotated on all the levels). Detailed annotation guidelines that constitute the basis of the handling with the language material were developed (Mikulov\u00e1 et al., 2005) based on the theoretical assumptions of the Functional Generative Grammar (for the first formulations of this formal framework, see Sgall, 1967; Sgall et al., 1986) . The annotation of the information structure in PDT is also based on this theory. The same linguistic approach was used in some other annotation schemes connected with the annotation of topic-focus articulation (e.g. Postolache, 2005) .", "cite_spans": [ { "start": 213, "end": 231, "text": "Haji\u010d et al., 2006", "ref_id": "BIBREF6" }, { "start": 232, "end": 264, "text": ", updated in Bej\u010dek et al., 2012", "ref_id": null }, { "start": 666, "end": 689, "text": "(Mikulov\u00e1 et al., 2005)", "ref_id": "BIBREF12" }, { "start": 822, "end": 834, "text": "Sgall, 1967;", "ref_id": "BIBREF16" }, { "start": 835, "end": 854, "text": "Sgall et al., 1986)", "ref_id": "BIBREF18" }, { "start": 1073, "end": 1090, "text": "Postolache, 2005)", "ref_id": "BIBREF15" } ], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Prague Treebanks", "sec_num": "1.1" }, { "text": "Our effort is concentrated on annotating the topic-focus articulation (TFA) in a parallel corpus -the Prague Czech-English Dependency Treebank (PCEDT), to make possible contrastive studies of this phenomenon. As the first step, we annotate topic-focus articulation in the Czech part of the treebank. The annotation guidelines have been taken over from the PDT approach, i.e. they also follow the theory of Functional Generative Description.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Aim of the Paper", "sec_num": "1.2" }, { "text": "In Section 2, we give an overview of the theoretical background of TFA, Section 3 introduces the Prague Czech-English Dependency Treebank (the data to be annotated). Section 4 describes in detail an automatic pre-annotation procedure that was applied on the data before they were annotated manually by a human annotator. The final step of this part of our research was the evaluation of effectiveness of the automatic pre-annotation, given in Section 5.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Aim of the Paper", "sec_num": "1.2" }, { "text": "The theoretical linguistic background for the creating of the whole corpus PCEDT is the Functional Generative Description (Sgall, 1967; Sgall et al., 1986) . Topic-focus articulation in this theoretical framework was described especially by Sgall and Haji\u010dov\u00e1 (summarized in Sgall et al., 1986, Haji\u010dov\u00e1 et al., 1998) . On the basis of this, the annotation guidelines for manual annotation of topic-focus articulation in the Prague Dependency Treebank (PDT) were established and are available in the annotation manual for the underlying structure of sentences in Mikulov\u00e1 et al. (2005) . These guidelines are used also for the Czech part of the Prague Czech-English Dependency Treebank.", "cite_spans": [ { "start": 122, "end": 135, "text": "(Sgall, 1967;", "ref_id": "BIBREF16" }, { "start": 136, "end": 155, "text": "Sgall et al., 1986)", "ref_id": "BIBREF18" }, { "start": 241, "end": 250, "text": "Sgall and", "ref_id": "BIBREF16" }, { "start": 251, "end": 317, "text": "Haji\u010dov\u00e1 (summarized in Sgall et al., 1986, Haji\u010dov\u00e1 et al., 1998)", "ref_id": null }, { "start": 563, "end": 585, "text": "Mikulov\u00e1 et al. (2005)", "ref_id": "BIBREF12" } ], "ref_spans": [], "eq_spans": [], "section": "Theoretical Background for Corpus Annotation of Topic-Focus Articulation in PCEDT", "sec_num": "2" }, { "text": "The theory of topic-focus articulation within the framework of Functional Generative Description is based on the aboutness-principle: the topic is the part of a sentence that is spoken about, and, complementarily, the focus is the sentence part that declares something about the topic. From the cognitive point of view, topic may be characterized as the \"given\" part of the sentence and focus as the \"new\" one. However, this does not mean that the focus elements cannot be mentioned in the previous language context at all but they have to bring some non-identifiable information or information in new relations. Most sentences contain both parts -topic and focus. However, some sentences can be contextually independent (e.g. the first sentence of the text or its title) and they do not have to contain the topic part (these are topic-less sentences). On the contrary, the focus is an obligatory component of every sentence -it is the informatively more important part of the message than the topic.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "The basic opposition established by the TFA theory and included in the annotation scheme is the opposition of contextual boundness: each element of the underlying structure of the sentence carries the feature \"contextually bound\" or \"contextually non-bound\". In addition, the contextually bound elements in the topic can be either contrastive, or non-contrastive. Contrastive contextually bound sentence members differ from the non-contrastive ones in the presence of a contrastive stress and in their semantic contentthey express contrast to some previous context (e.g. at home -abroad).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "Non-contrastive contextually bound expressions are marked as 't', contrastive contextually bound expressions are marked as 'c' and contextually non-bound expressions are marked as 'f' 2 .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "The opposition between contextually bound and contextually non-bound elements serves then as a basis for the bi-partition of the sentence into its topic and focus; according to this hypothesis, an algorithm for topic-focus bi-partition was formulated, implemented and tested on the PDT data, with some rather encouraging results (see Haji\u010dov\u00e1 et al., 2005) .", "cite_spans": [ { "start": 334, "end": 356, "text": "Haji\u010dov\u00e1 et al., 2005)", "ref_id": "BIBREF8" } ], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "In Czech (Czech is the language of Prague Dependency Treebank and also of one half of the Prague Czech-English Dependency Treebank), the word order position of predicative verb is often the natural boundary between the topic and focus part in the sentence -cf. Example (1).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "( 1) Several operational tests have been proposed in literature that help to distinguish between topic and focus, the most relevant of them being the question test and the test of negation (for details see Sgall et al., 1986; Haji\u010dov\u00e1 et al., 1998) .", "cite_spans": [ { "start": 206, "end": 225, "text": "Sgall et al., 1986;", "ref_id": "BIBREF18" }, { "start": 226, "end": 248, "text": "Haji\u010dov\u00e1 et al., 1998)", "ref_id": "BIBREF10" } ], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "In short, the basis of the question test is to ask a question that fully represents the context for the tested sentence. The tested sentence has to be a relevant answer to the question. The sentence members present in both the question and answer are topic members. The elements present only in the answer are members of the focus.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "The principle of the negation test is to find out the possible scope of negation in the negative counterpart to the given sentence. In principle, the sentence members that are in the scope of negation in the given context belong to the focus part of the sentence. Other members form the topic part. However, there is a possibility of negative topic, i.e. the topic of the sentence is negated and the focus stands out of the scope (for details see e.g. Sgall et al., 1973) .", "cite_spans": [ { "start": 452, "end": 471, "text": "Sgall et al., 1973)", "ref_id": "BIBREF17" } ], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "For detailed information on annotation guidelines of topic-focus articulation in the framework of Functional Generative Description, the online annotation manual is available (see http://ufal.mff.cuni.cz/pdt2.0/doc/manuals/en/tlayer/html/index.html).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Topic-Focus Articulation in Functional Generative Description", "sec_num": "2.1" }, { "text": "The annotation effort described in this paper is performed on data from the Prague Czech-English Dependency Treebank (PCEDT, Haji\u010d et al., 2012) , a manually parsed parallel Czech-English corpus that contains over 1.2 million running words (50 thousand sentences in each of the two languages). The English part consists of texts from the Penn Treebank (Marcus et al., 1993) articles from the Wall Street Journal. The Czech part contains human translations of the English sentences to Czech. The annotation (on both language sides) is performed on four language layers: the \"word\" layer, the morphological layer, the analytical layer (i.e. the layer of surface syntax) and the tectogrammatical layer (i.e. the semantic layer of the deep syntax).", "cite_spans": [ { "start": 125, "end": 144, "text": "Haji\u010d et al., 2012)", "ref_id": "BIBREF7" }, { "start": 352, "end": 373, "text": "(Marcus et al., 1993)", "ref_id": "BIBREF11" } ], "ref_spans": [], "eq_spans": [], "section": "Language Material -Prague Czech-English Dependency Treebank", "sec_num": "3" }, { "text": "On the topmost (tectogrammatical) layer, individual sentences are organized in dependency tree structures, according to the style of the Prague Dependency Treebank (PDT). Autosemantic words and coordinating structures are captured in the trees, as well as the valency of verbs (each language has its own valency lexicon in PCEDT). Additionally, the surface sentence ellipsis is reconstructed in the deep sentence structure and also pronominal anaphoric relations are labeled in the texts. The topic-focus articulation is also to be annotated on this layer.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Language Material -Prague Czech-English Dependency Treebank", "sec_num": "3" }, { "text": "The parallel Czech-English data are aligned manually on the level of sentences and automatically on the level of tectogrammatical nodes.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Language Material -Prague Czech-English Dependency Treebank", "sec_num": "3" }, { "text": "More detailed information on PCEDT is available on the project website (http://ufal.mff.cuni.cz/pcedt2.0/en/index.html).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Language Material -Prague Czech-English Dependency Treebank", "sec_num": "3" }, { "text": "For the annotation of topic-focus articulation in the Czech part of PCEDT, an automatic pre-annotation procedure was developed. The particular steps (rules) of the pre-annotation were mainly established on the basis of the completed annotation of contextual boundness in the Prague Dependency Treebank (i.e. on the basis of annotated Czech texts). The cross-language alignment of tectogrammatical nodes in PCEDT was also exploited (see the pre-annotation step 10 below), allowing for taking advantage of the existence of indefinite articles in English (not present in the Czech language).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Automatic Pre-Annotation", "sec_num": "4" }, { "text": "Using information from the English side for the pre-annotation of topic-focus articulation in the Czech part is possible, as the topic-focus articulation of the given sentence in the given context should be identical regardless on the language 4 . The surface word order may vary in Czech in comparison with English (cf. the different word order in Example (1) in the two languages) but the topic-focus articulation of the sentence should be the same in both the languages. This theoretical assumption, as well as the quality of the English->Czech translation (from the point of view of topic-focus articulation), can be tested on real corpus data once the annotation on both language sides of PCEDT is finished.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Automatic Pre-Annotation", "sec_num": "4" }, { "text": "So far, the automatic procedure was used for pre-annotation of a sample of the PCEDT Czech part and this pre-annotated sample was subsequently manually annotated by a human annotator. The annotator checked the correctness of the pre-annotation and annotated the rest of the nodes (nodes that had not been pre-annotated). Afterwards, it was evaluated how many changes of the automatic pre-annotation of topic-focus articulation the human annotator had to carry out, i.e. how many mistakes the automatic pre-annotation had made in the data.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Automatic Pre-Annotation", "sec_num": "4" }, { "text": "It should be noted that the goal of the automatic pre-annotation was to help the human annotators with simple decisions, not to classify every sentence member as contextually bound ('t') or non-bound ('f') element. Our intention was to apply only reliable rules and leave too complex decisions (often depending on the meaning of the text) on the human annotator. We wanted to avoid introducing too many errors in the pre-annotation, as human annotators might be prone to overlooking errors in already annotated nodes and concentrate only (or at least better) on the so far unannotated nodes. For the selection of the pre-annotation steps, we estimated their expected error rates (where possible) based on measurements on the topic-focus annotation in PDT (see the expected error rates of the individual pre-annotation steps below in 4.1). For using a rule, we set the maximum number of expected errors to 10 %.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Automatic Pre-Annotation", "sec_num": "4" }, { "text": "The following steps have been performed during the automatic pre-annotation. For each step (where possible), we give an estimate of the preannotation error (expected error rate, EER), based on the measurement of the phenomenon in the data of Prague Dependency Treebank. The steps have been applied in the presented order.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "Step 10 takes advantage of the cross-language alignment of words in PCEDT.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "1. Nodes generated on the tectogrammatical layer without a counterpart on the analytical layer (i.e. newly added, but not copied nodes in the tectogrammatical representation) and that do not have functor=RHEM (rhematizer), nor t_lemma=#Forn (part of a phrase in a foreign language), get automatically assigned tfa='t', i.e. contextually bound, (EER: 0). For an example, see Figure 1 . 5 Figure 1 represents the following Czech sentence -Example (2) from PCEDT:", "cite_spans": [], "ref_spans": [ { "start": 374, "end": 382, "text": "Figure 1", "ref_id": null }, { "start": 387, "end": 395, "text": "Figure 1", "ref_id": null } ], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "(2) \"Pro\u010d David Dinkins,\" \u0159\u00edk\u00e1 kri tik, \"v\u017edycky vy\u010dk\u00e1v\u00e1, dokud nen\u00ed chycen p\u0159i \u010dinu?\" \"David Dinkins,\" says the kicker, \"Why does he always wait until he's caught?\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "In the surface (analytical) structure of the given sentence with the Czech verb \u0159\u00edkat (to say), the Addressee is not present explicitly although this verb has the Addressee (apart from the Effect, the Actor and the non-obligatory Patient) in its valency frame (someone.obligatory_Actor says some thing.obligatory_Effect to someone.obligatory_Addressee about something/somebody.non-obligatory_Patient). So the Addressee is present only in the deep (tectogrammatical) sentence structure (in Figure 1 , it is captured as a small square with the symbol of Addressee ADDR). The sentence members that appear only implicitly in the sentence (as the Addressee in this case) are not supposed to carry some new, important information (because their presence in the /surface part of the/ sentence is not necessary) and therefore they are automatically pre-annotated as contextually bound (furand the tectogrammatical layer) are displayed as small circles in the figure. Members that are present only in the deep sentence structure (on the tectogrammatical layer) and do not appear in the surface sentence structure (i.e. not on the analytical layer) are displayed as small squares. White colour represents contextually bound sentence members (they are also depicted with 't' next to the lemma); yellow colour (light grey in b/w) represents contextually non-bound sentence members (they are depicted with 'f'). The grey members do not have any value of contextual boundness yet (they were not automatically pre-annotated and they will be manually annotated by a human annotator). ther examples are the sentence members Patient PAT and Actor ACT by the Czech verb chytitto catch: somebody.obligatory_Actor catches some one.obligatory_Patient, see Figure 1 ).", "cite_spans": [], "ref_spans": [ { "start": 489, "end": 497, "text": "Figure 1", "ref_id": null }, { "start": 1734, "end": 1742, "text": "Figure 1", "ref_id": null } ], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "2. Nodes generated at the tectogrammatical layer that are members of coordination/apposition and have an analytical counterpart (they are copied nodes; it also means that it is not e.g. #Forn), get assigned tfa='t', i.e. contextually bound, (EER: 0), see Example 3from PCEDT.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "(3) \"Nyn\u00ed,\" \u0159\u00edk\u00e1 Joseph Napolitan, pr\u016fkopn\u00edk politick\u00e9 televize, \"je c\u00edlem j\u00edt do \u00fatoku jako prvn\u00ed, po sledn\u00ed a [j\u00edt] t v\u017edycky.\" \"Now,\" says Joseph Napolitan, a pio neer in political television, \"the idea is to attack first and [to attack] t always.\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "This pre-annotation step concerns also other cases of sentence members that are not present in the surface (analytical) structure but appear in the deep (tectogrammatical) layer. These nodes are not newly added to the structure, e.g. because of the valency verb frame, but they appeared in some previous structures and they are omitted in the surface structure (and copied to the deep structure) because the reader can understand them easily from the previous context as in the phrases from Example (3): to attack first and (to attack) always. Since these members (present only implicitly in the sentence) are obviously deducible from the context, they are considered as contextually bound and therefore they are pre-annotated as such.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "3. Nodes where a grammatical, textual or segment coreference starts, get tfa='t', i.e. contextually bound, (EER: 1:100), see Example (4) from PCEDT. This step of the automatic pre-annotation takes advantage of the finished annotation of coreference in the PCEDT texts. Sentence elements that are anaphors 6 of a coreference relation are sup-posed to be contextually bound and therefore they are automatically assigned the value 't'.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "There are two coreference relations in Example (4): 1. Dinkins -sv\u00fdch (he); 2. mu\u017e (man)kter\u00e9ho (his). The members that refer to some previous sentence members (sv\u00fdch and kter\u00e9ho in this case) are automatically pre-annotated as contextually bound.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "In another example from PCEDT, depicted in Figure 2 , starting nodes (anaphors) of grammatical coreference (three intra-sentential more or less vertical arrows) and textual coreference (two horizontal arrows going from the second tree to the first one) are pre-annotated as contextually bound.", "cite_spans": [], "ref_spans": [ { "start": 43, "end": 51, "text": "Figure 2", "ref_id": null } ], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": "4. Nodes with functor=PRED that are not newly generated and whose t_lemma does not appear in the previous sentence, get tfa='f', i.e. contextually non-bound, (EER: 1:40), see Example (5) from PCEDT. The data of previously annotated Prague Dependency Treebank demonstrated that most Predicates (in corpus marked as PRED) are contextually non-bound -therefore, they are pre-annotated as 'f'.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Steps of the Pre-Annotation", "sec_num": "4.1" }, { "text": ". Newly generated nodes with functor=PRED get tfa='t', i.e. contextually bound, (EER: 1:100), see Example (6) from PCEDT.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "In contrast to the step 4), Predicates that are not present in the surface sentence structure are preannotated as contextually bound, cf. step 3). 6. Other verbal nodes (gram/sempos=v) with functor from the set {ADDR, AIM, CAUS, ACMP, MANN, PAT, EFF, AUTH, BEN, COMPL, EXT, ORIG, RESL, TFHL, TSIN} get tfa='f', i.e. contextually non-bound, (EER: 1:10), see Example (7) from PCEDT.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "anaphor (the latter in the text) and antecedent (the former) are connected by a coreference relation.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "The data of the Prague Dependency Treebank also demonstrated that most sentence members expressed as dependent clauses (i.e. containing a finite verb) and having the semantic role of Addressee, Aim, Cause, Accompaniment, Patient, Effect, Author, Benefactor, Complement, Extent, Origo, Result or Temporal modifications (expressing for how long or since when) are contextually non-bound -therefore, they are pre-annotated as non-bound also in data of the Prague Czech-English Dependency Treebank. 7. Nodes with functor from the set {PARTL, DENOM, MOD, EXT} get tfa='f', i.e. contextually non-bound, (EER: 1:10), see again Example (6) above from PCEDT.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "The data of the Prague Dependency Treebank further demonstrated that most sentence members assigned the semantic role of independent interjectional clause (marked as PARTL), independent non-parenthetical nominal clause (DE-NOM), atomic expression with a modal meaning (MOD) or adjunct expressing extent (EXT) are contextually non-bound and therefore they are pre-annotated as such.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "In the Example (6), the sentence member pravd\u011bpodobn\u011b (presumably) is in the role of an atomic expression with a modal meaning (MOD) and therefore it will be automatically assigned the value 'f'. 8. Nodes with functor=RHEM (i.e. they have a function of a rhematizer) that are not in the first position in the sentence, get tfa='f', i.e. contextually non-bound, (EER: 1:10), see Example (8) from PCEDT. The rhematizers (as e.g. English particles only, for example, also, especially, principally) mostly precede a focus element and in the theory of TFA, they are also considered contextually nonbound. However, also contrastive contextually bound expressions can follow the rhematizerstypically at the beginning of the sentence (and in this case, also the rhematizers are contextually bound). Therefore, only such rhematizers are pre-annotated as contextually non-bound that are not placed in the initial position in the sentence. 9. Nodes with t_lemma=tady (here) get tfa='t', i.e. contextually bound, (EER: 1:10), see Example (9) from PCEDT.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "Some lemmas (especially with a deictic function like here) appear as contextually bound in most cases (but not in all -see e.g. What happens heref and now?), which observation is also made use of in the automatic pre-annotation. 10. Nodes that are Czech counterparts of English nodes that in the English sentence are placed after their governing verb on the surface and that are preceded by an indefinite article, get tfa='f', i.e. contextually nonbound, (EER: unknown), see Example 10from PCEDT. In Example (10), the sentence member victim is modified by the indefinite article a in the English variant of the sentence, which leads to the assumption that this member is contextually nonbound. Since the value of the same sentence member should be identical both in English and in Czech variant of the sentence, also the Czech member ob\u011b\u0165 (that is the counterpart of the vic tim) is supposed to be contextually non-bound.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "The following steps of the automatic pre-annotation are performed after the previous steps have been applied on all nodes of the given tree:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "11. Daughters of a verb that has tfa='f' and that is not on the first or second position (in its clause), if they appear after the governing verb on the surface, get tfa='f', i.e. contextually non-bound, (EER: unknown), see Example (11) from PCEDT. This step of the pre-annotation makes use of the fact that in Czech, the surface word order often is used to express the topic-focus articulation. Under the condition that the contextually non-bound predicative verb is placed further to the right than on the second position in the sentence and that the sentence has a non-marked word order 7 (i.e. emotionally neutral), it is possible to assume that the sentence members following the predicative verb are contextually non-bound.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "12. Nodes with functor=RSTR that are daughters of a node with tfa='f', get tfa='f', i.e. contextually non-bound, (EER: 1:30). The final step of the automatic pre-annotation is based on the fact that the adnominal adjuncts modifying its governing noun (in the annotated corpus marked as RSTR) often have a very high degree of communicative dynamism because their primary function is to specify something. Therefore, they are pre-annotated as contextually bound (if they modify a non-bound element at the same time).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "5", "sec_num": null }, { "text": "At the time of submitting the final version of the paper, more than one thousand automatically pre-annotated sentences have also been manually annotated by a human annotator 8 and could be used for evaluation of the pre-annotation. In 59 documents (1,145 sentences, 22,436 nodes on the tectogrammatical layer), 7,864 nodes out of 19,105 tfa-relevant nodes have been automatically pre-annotated (i.e. 41.1 %). Table 1 gives an overview of how many times the individual pre-annotation steps have been applied. Based on the estimates presented in Sec-tion 4.1 (for the two unknown estimates in steps 10 and 11 we used EER: 1:10), we can calculate the expected number of errors in the pre-annotation as (about) 340 errors. In the manual annotation, the annotator changed the pre-annotated value in 294 cases (i.e. 3.7 % of pre-annotated nodes). Table 2 shows details on the manually performed changes. The numbers show that the automatic pre-annotation is more successful in marking contextually bound sentence members, as only 0.8 % of nodes pre-annotated as 't' and 5.4 % of nodes pre-annotated as 'f' were manually changed to another value.", "cite_spans": [], "ref_spans": [ { "start": 409, "end": 416, "text": "Table 1", "ref_id": "TABREF8" }, { "start": 841, "end": 848, "text": "Table 2", "ref_id": "TABREF9" } ], "eq_spans": [], "section": "Evaluation of the Automatic Pre-Annotation", "sec_num": "5" }, { "text": "PDT 2.0 sample of PCEDT contr. contextually bound ('c') 5.4 % 5.7 % non-contr. contextually bound ('t')", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Evaluation of the Automatic Pre-Annotation", "sec_num": "5" }, { "text": "31.3 % 33.6 % contextually nonbound ('f') 63.3 % 60.7 % Table 3 : The percentage distribution of manually annotated TFA-values in PDT (training data) and so far annotated sample of the Czech part of PCEDT", "cite_spans": [], "ref_spans": [ { "start": 56, "end": 63, "text": "Table 3", "ref_id": null } ], "eq_spans": [], "section": "Evaluation of the Automatic Pre-Annotation", "sec_num": "5" }, { "text": "The inability of the pre-annotation procedure to set the 'c' value (contrastive contextually bound) does not harm the results much, as only 37 (0.5 %) pre-annotated nodes were manually changed to this value, and the overall ratio of contrastive contextually bound nodes among all (manually) annotated nodes both in PDT and PCEDT is less than 6 % (see Table 3 ). The main limitations of the pre-annotation are in its coverage (more than half of the nodes are not pre-annotated) and in its natural inability to take the meaning of the text into account (and thus being unable to better distinguish between 't' and 'f' values).", "cite_spans": [], "ref_spans": [ { "start": 351, "end": 358, "text": "Table 3", "ref_id": null } ], "eq_spans": [], "section": "Evaluation of the Automatic Pre-Annotation", "sec_num": "5" }, { "text": "From another point of view, the results suggest that the expected error rates (estimated on PDT) are accurate and that the automatic pre-annotation is sufficiently reliable and serves as a substantial help to the annotators. 9", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Evaluation of the Automatic Pre-Annotation", "sec_num": "5" }, { "text": "The paper presented the first part of the project of parallel annotation of topic-focus articulation in the Prague Czech-English Dependency Treebank (PCEDT). We described the annotation principles and schemes, and elaborated on 12 automatic steps of the pre-annotation procedure for the Czech part of the treebank. The pre-annotation is able to mark over 40 % of the whole text (the rest is supposed to be annotated by human annotators). It can distinguish between contextually bound and non-bound sentence elements with the average success rate over 96 %, as shown by the evaluation on manually annotated texts.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "6" }, { "text": "Given the available funds, our present goal is to annotate 5 thousand parallel sentences.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The contextually non-bound elements do not have a contrastive and non-contrastive variant in the theory of FGP.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The members that carry the centre of the intonation in the sentence are capitalized (in the translation).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "In fact, the topic-focus articulation of the given sentence is the same regardless on the language. However, we operate with a parallel corpus -the English part contains original texts and the Czech one their translations. It is possible that the Czech translations could be inaccurate in some casesespecially regarding the topic-focus articulation. Therefore, the value of contextual boundness could differ in both parts of parallel corpus in a few cases.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Sentence members (nodes) that are really expressed in the surface sentence structure (that appear on both the analytical", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "A reference to an entity or event that has already been mentioned in the preceding text; the two mentions -", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The human annotator decides whether the word order is marked or non-marked (it is not possible to check it automatically in our procedure of pre-annotation). 8 There were actually two annotators, working on different parts of the data. For simplicity, we refer to them as 'a human annotator'. Only during a training phase (performed on a few documents), the two annotators worked on the same data and their discrepancies were subsequently checked by an arbiter and discussed.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Of course, it is a matter of discussion (and testing), how much effort of the human annotator such a pre-annotation saves and how to set the reliability limit for the rule selection.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [ { "text": "We gratefully acknowledge support from the Grant Agency of the Czech Republic (grants P406/12/0658 and P406/2010/0875). This work has been using language resources developed and/or stored and/or distributed by the LINDAT-Clarin project of the Ministry of Education of the Czech Republic (project LM2010013).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Acknowledgment", "sec_num": null } ], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Prague Dependency Treebank 2.5 -a revisited version of PDT 2.0", "authors": [ { "first": "E", "middle": [], "last": "Bej\u010dek", "suffix": "" }, { "first": "J", "middle": [], "last": "Panevov\u00e1", "suffix": "" }, { "first": "J", "middle": [], "last": "Popelka", "suffix": "" }, { "first": "P", "middle": [], "last": "Stra\u0148\u00e1k", "suffix": "" }, { "first": "M", "middle": [], "last": "\u0160ev\u010d\u00edkov\u00e1", "suffix": "" }, { "first": "J", "middle": [], "last": "\u0160t\u011bp\u00e1nek", "suffix": "" }, { "first": "Z", "middle": [], "last": "\u017dabokrtsk\u00fd", "suffix": "" } ], "year": 2012, "venue": "Proceedings of the 24th Inter national Conference on Computational Linguistics", "volume": "", "issue": "", "pages": "231--246", "other_ids": {}, "num": null, "urls": [], "raw_text": "E. Bej\u010dek, J. Panevov\u00e1, J. Popelka, P. Stra\u0148\u00e1k, M. \u0160ev\u010d\u00edkov\u00e1, J. \u0160t\u011bp\u00e1nek, Z. \u017dabokrtsk\u00fd. 2012. Prague Dependency Treebank 2.5 -a revisited ver- sion of PDT 2.0. In: Proceedings of the 24th Inter national Conference on Computational Linguistics (Coling 2012), Mumbai, India, pp. 231-246.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "A Framework for Annotating Information Structure in Discourse", "authors": [ { "first": "S", "middle": [], "last": "Calhoun", "suffix": "" }, { "first": "M", "middle": [], "last": "Nissim", "suffix": "" }, { "first": "M", "middle": [], "last": "Steedman", "suffix": "" }, { "first": "J", "middle": [], "last": "Brenier", "suffix": "" } ], "year": 2005, "venue": "Proceedings of the Workshop on Frontiers in Corpus Annotations II: Pie in the Sky", "volume": "", "issue": "", "pages": "45--52", "other_ids": {}, "num": null, "urls": [], "raw_text": "S. Calhoun, M. Nissim, M. Steedman, J. Brenier. 2005. A Framework for Annotating Information Structure in Discourse. In: Proceedings of the Workshop on Frontiers in Corpus Annotations II: Pie in the Sky. Ann Arbor, Michigan: Association for Computational Linguistics, pp. 45-52. URL http://aclweb.org/anthology/W/W05/W05-0307.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Annotating information structure. The case of \"topic", "authors": [ { "first": "P", "middle": [], "last": "Cook", "suffix": "" }, { "first": "F", "middle": [], "last": "Bildhauer", "suffix": "" } ], "year": 2011, "venue": "Beyond Semantics.Corpus based Investigations of Pragmatic and Discourse Phenomena", "volume": "", "issue": "", "pages": "45--56", "other_ids": {}, "num": null, "urls": [], "raw_text": "P. Cook, F. Bildhauer. 2011. Annotating information structure. The case of \"topic\". In: S. Dipper & H. Zinsmeister (eds.), Beyond Semantics.Corpus based Investigations of Pragmatic and Discourse Phenomena, Ruhr Universit\u00e4t Bochum, Bochumer Linguistische Arbeitsberichte, pp. 45-56. URL http://www.linguistics.ruhr-uni-bochum.de/bla/be- yondsem2011/cook_final.pdf.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Infor mation Structure in CrossLinguistic Corpora: An notation Guidelines for Phonology", "authors": [], "year": 2007, "venue": "of Interdisciplinary Studies on Information Structure", "volume": "7", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "S. Dipper, M. G\u00f6tze, S. Skopeteas (eds.). 2007. Infor mation Structure in CrossLinguistic Corpora: An notation Guidelines for Phonology, Morphology, Syntax, Semantics and Information Structure, vol. 7 of Interdisciplinary Studies on Information Struc- ture. Potsdam, Germany: Universit\u00e4tsverlag Pots- dam. URL http://www.sfb632.uni-potsdam.de/pub- lications/isis07.pdf.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "AN-NIS: A Linguistic Database for Exploring Information Structure", "authors": [ { "first": "S", "middle": [], "last": "Dipper", "suffix": "" }, { "first": "M", "middle": [], "last": "G\u00f6tze", "suffix": "" }, { "first": "M", "middle": [], "last": "Stede", "suffix": "" }, { "first": "T", "middle": [], "last": "Wegst", "suffix": "" } ], "year": 2004, "venue": "Interdisciplinary Studies on Information Structure (ISIS)", "volume": "1", "issue": "", "pages": "245--279", "other_ids": {}, "num": null, "urls": [], "raw_text": "S. Dipper, M. G\u00f6tze, M. Stede, T. Wegst. 2004. AN- NIS: A Linguistic Database for Exploring Informa- tion Structure. In Ishihara, S., Schmitz, M., Schwarz, A. (Eds.), Working Papers of the SFB632, Interdisciplinary Studies on Information Structure (ISIS) 1, pp. 245-279. Potsdam: Univer- sity publishing house Potsdam.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Zur informationsstrukturellen Annotation sprachhistorischer Texten. Sprache und Informationsverarbeitung 31", "authors": [ { "first": "K", "middle": [], "last": "Donhauser", "suffix": "" } ], "year": 2007, "venue": "", "volume": "", "issue": "", "pages": "39--45", "other_ids": {}, "num": null, "urls": [], "raw_text": "K. Donhauser. 2007. Zur informationsstrukturellen Annotation sprachhistorischer Texten. Sprache und Informationsverarbeitung 31, pp. 39-45. URL http://www.sfb632.uni- potsdam.de/publications/B4/donhauser_2007.pdf.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Prague Dependency Treebank 2.0. Software prototype, Linguistic Data Consortium", "authors": [ { "first": "J", "middle": [], "last": "Haji\u010d", "suffix": "" }, { "first": "J", "middle": [], "last": "Panevov\u00e1", "suffix": "" }, { "first": "E", "middle": [], "last": "Haji\u010dov\u00e1", "suffix": "" }, { "first": "P", "middle": [], "last": "Sgall", "suffix": "" }, { "first": "P", "middle": [], "last": "Pajas", "suffix": "" }, { "first": "J", "middle": [], "last": "\u0160t\u011bp\u00e1nek", "suffix": "" }, { "first": "J", "middle": [], "last": "Havelka", "suffix": "" }, { "first": "M", "middle": [], "last": "Mikulov\u00e1", "suffix": "" }, { "first": "Z", "middle": [], "last": "\u017dabokrtsk\u00fd", "suffix": "" }, { "first": "M", "middle": [], "last": "\u0160ev\u010d\u00edkov\u00e1-Raz\u00edmov\u00e1", "suffix": "" } ], "year": 2006, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "J. Haji\u010d, J. Panevov\u00e1, E. Haji\u010dov\u00e1, P. Sgall, P. Pajas, J. \u0160t\u011bp\u00e1nek, J. Havelka, M. Mikulov\u00e1, Z. \u017dabokrtsk\u00fd, M. \u0160ev\u010d\u00edkov\u00e1-Raz\u00edmov\u00e1. 2006. Prague Dependency Treebank 2.0. Software proto- type, Linguistic Data Consortium, Philadelphia, PA, USA, ISBN 1-58563-370-4, http://www.ldc.u- penn.edu, Jul 2006.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Announcing Prague Czech-English Dependency Treebank 2.0", "authors": [ { "first": "J", "middle": [], "last": "Haji\u010d", "suffix": "" }, { "first": "E", "middle": [], "last": "Haji\u010dov\u00e1", "suffix": "" }, { "first": "J", "middle": [], "last": "Panevov\u00e1", "suffix": "" }, { "first": "P", "middle": [], "last": "Sgall", "suffix": "" }, { "first": "O", "middle": [], "last": "Bojar", "suffix": "" }, { "first": "S", "middle": [], "last": "Cinkov\u00e1", "suffix": "" }, { "first": "E", "middle": [], "last": "Fu\u010d\u00edkov\u00e1", "suffix": "" }, { "first": "M", "middle": [], "last": "Mikulov\u00e1", "suffix": "" }, { "first": "P", "middle": [], "last": "Pajas", "suffix": "" }, { "first": "J", "middle": [], "last": "Popelka", "suffix": "" }, { "first": "J", "middle": [], "last": "Semeck\u00fd", "suffix": "" }, { "first": "J", "middle": [], "last": "\u0160indlerov\u00e1", "suffix": "" }, { "first": "J", "middle": [], "last": "\u0160t\u011bp\u00e1nek", "suffix": "" }, { "first": "J", "middle": [], "last": "Toman", "suffix": "" }, { "first": "Z", "middle": [], "last": "Ure\u0161ov\u00e1", "suffix": "" }, { "first": "Z", "middle": [], "last": "\u017dabokrtsk\u00fd", "suffix": "" } ], "year": 2012, "venue": "Proceedings of the 8th International Conference on Language Resources and Evalua tion (LREC 2012)", "volume": "", "issue": "", "pages": "3153--3160", "other_ids": {}, "num": null, "urls": [], "raw_text": "J. Haji\u010d, E. Haji\u010dov\u00e1, J. Panevov\u00e1, P. Sgall, O. Bojar, S. Cinkov\u00e1, E. Fu\u010d\u00edkov\u00e1, M. Mikulov\u00e1, P. Pajas, J. Popelka, J. Semeck\u00fd, J. \u0160indlerov\u00e1, J. \u0160t\u011bp\u00e1nek, J. Toman, Z. Ure\u0161ov\u00e1, Z. \u017dabokrtsk\u00fd. 2012. An- nouncing Prague Czech-English Dependency Tree- bank 2.0. In: Proceedings of the 8th International Conference on Language Resources and Evalua tion (LREC 2012), European Language Resources Association, \u0130stanbul, Turkey, ISBN 978-2- 9517408-7-7, pp. 3153-3160.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Corpus Evidence of Contextual Boundness and Focus", "authors": [ { "first": "E", "middle": [], "last": "Haji\u010dov\u00e1", "suffix": "" }, { "first": "J", "middle": [], "last": "Havelka", "suffix": "" }, { "first": "K", "middle": [], "last": "Vesel\u00e1", "suffix": "" } ], "year": 2005, "venue": "Proceedings of the Corpus Linguistics Conference Series", "volume": "", "issue": "", "pages": "1747--9398", "other_ids": {}, "num": null, "urls": [], "raw_text": "E. Haji\u010dov\u00e1, J. Havelka, K. Vesel\u00e1. 2005. Corpus Evi- dence of Contextual Boundness and Focus. In: Proceedings of the Corpus Linguistics Conference Series, University of Birmingham, Birmingham, UK, ISSN 1747-9398.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "A Manual for Tectogrammatical Tagging of the Prague Dependency Treebank", "authors": [ { "first": "E", "middle": [], "last": "Haji\u010dov\u00e1", "suffix": "" }, { "first": "J", "middle": [], "last": "Panevov\u00e1", "suffix": "" }, { "first": "P", "middle": [], "last": "Sgall", "suffix": "" } ], "year": 2000, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "E. Haji\u010dov\u00e1, J. Panevov\u00e1, P. Sgall. 2000. A Manual for Tectogrammatical Tagging of the Prague De- pendency Treebank. Technical report tr200009, \u00daFAL/CKL. URL http://ufal.mff.cuni.cz/pdt/Cor- pora/PDT_1.0/Doc/tmanual/tmanen.pdf. In cooper- ation with A. B\u00f6hmov\u00e1, M. Ceplov\u00e1 and V. \u0158ezn\u00ed\u010dkov\u00e1. Translated by Z. Kirschner, E. Ha- ji\u010dov\u00e1 and P. Sgall.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Topicfocus articulation, tripartite structures, and semantic content", "authors": [ { "first": "E", "middle": [], "last": "Haji\u010dov\u00e1", "suffix": "" }, { "first": "B", "middle": [ "H" ], "last": "Partee", "suffix": "" }, { "first": "P", "middle": [], "last": "Sgall", "suffix": "" } ], "year": 1998, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "E. Haji\u010dov\u00e1, B. H. Partee, P. Sgall. 1998. Topicfocus articulation, tripartite structures, and semantic content. Dordrecht, Boston: Kluwer Academic Publishers.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "Building a large annotated corpus of English: The Penn Treebank", "authors": [ { "first": "M", "middle": [ "P" ], "last": "Marcus", "suffix": "" }, { "first": "B", "middle": [], "last": "Santorini", "suffix": "" }, { "first": "M", "middle": [ "A" ], "last": "Marcinkiewicz", "suffix": "" } ], "year": 1993, "venue": "Computational Linguistics", "volume": "19", "issue": "2", "pages": "313--330", "other_ids": {}, "num": null, "urls": [], "raw_text": "M. P. Marcus, B. Santorini, M. A. Marcinkiewicz. 1993. Building a large annotated corpus of English: The Penn Treebank. Computational Linguistics, 19(2), pp. 313-330.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "Annotation on the tec togrammatical layer in the Prague Dependency Treebank. The Annotation Guidelines. Prague: UFAL MFF", "authors": [ { "first": "M", "middle": [], "last": "Mikulov\u00e1", "suffix": "" } ], "year": 2005, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "M. Mikulov\u00e1 et al. 2005. Annotation on the tec togrammatical layer in the Prague Dependency Treebank. The Annotation Guidelines. Prague: UFAL MFF. Available at: http://ufal.mff.cuni.cz/pdt2", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "An annotation scheme for information status in dialogue", "authors": [ { "first": "M", "middle": [], "last": "Nissim", "suffix": "" }, { "first": "S", "middle": [], "last": "Dingare", "suffix": "" }, { "first": "J", "middle": [], "last": "Carletta", "suffix": "" }, { "first": "M", "middle": [], "last": "Steedman", "suffix": "" } ], "year": 2004, "venue": "Proceedings of the 4thConference on Language Resources and Evaluation", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "M. Nissim, S. Dingare, J. Carletta, M. Steedman. 2004. An annotation scheme for information status in dialogue. In: Proceedings of the 4thConference on Language Resources and Evaluation. Lisbon, Portugal. URL http://www.lrec-conf.org/proceed- ings/lrec2004/pdf/638.pdf.", "links": null }, "BIBREF14": { "ref_id": "b14", "title": "Annotating Information Structure in a Corpus of Spoken Danish", "authors": [ { "first": "P", "middle": [], "last": "Paggio", "suffix": "" } ], "year": 2006, "venue": "Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC)", "volume": "", "issue": "", "pages": "1606--1609", "other_ids": {}, "num": null, "urls": [], "raw_text": "P. Paggio. Annotating Information Structure in a Cor- pus of Spoken Danish. 2006. In: Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC), pp. 1606-1609, Genoa, Italy.", "links": null }, "BIBREF15": { "ref_id": "b15", "title": "Learning Information Structure in The Prague Treebank", "authors": [ { "first": "O", "middle": [], "last": "Postolache", "suffix": "" } ], "year": 2005, "venue": "Proceedings of the ACL Student Research Workshop", "volume": "", "issue": "", "pages": "115--120", "other_ids": {}, "num": null, "urls": [], "raw_text": "O. Postolache. 2005. Learning Information Structure in The Prague Treebank. In: Proceedings of the ACL Student Research Workshop, pp. 115-120, Ann Arbor, Michigan, June 2005.", "links": null }, "BIBREF16": { "ref_id": "b16", "title": "Generative description of language and the Czech Declension (in Czech)", "authors": [ { "first": "P", "middle": [], "last": "Sgall", "suffix": "" } ], "year": 1967, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "P. Sgall. 1967. Generative description of language and the Czech Declension (in Czech). Prague: Academia.", "links": null }, "BIBREF17": { "ref_id": "b17", "title": "Topic, focus and generative semantics", "authors": [ { "first": "P", "middle": [], "last": "Sgall", "suffix": "" }, { "first": "E", "middle": [], "last": "Haji\u010dov\u00e1", "suffix": "" }, { "first": "E", "middle": [], "last": "Bene\u0161ov\u00e1", "suffix": "" } ], "year": 1973, "venue": "", "volume": "1", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "P. Sgall, E. Haji\u010dov\u00e1, E. Bene\u0161ov\u00e1. 1973. Topic, focus and generative semantics (Vol. 1). Kronberg Taunus: Scriptor Verlag.", "links": null }, "BIBREF18": { "ref_id": "b18", "title": "The meaning of the sentence in its semantic and pragmatic aspects", "authors": [ { "first": "P", "middle": [], "last": "Sgall", "suffix": "" }, { "first": "E", "middle": [], "last": "Haji\u010dov\u00e1", "suffix": "" }, { "first": "J", "middle": [], "last": "Panevov\u00e1", "suffix": "" } ], "year": 1986, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "P. Sgall, E. Haji\u010dov\u00e1, J. Panevov\u00e1. 1986. The meaning of the sentence in its semantic and pragmatic aspects. Springer.", "links": null } }, "ref_entries": { "FIGREF0": { "num": null, "type_str": "figure", "uris": null, "text": "\"Pamatujete si na Pinocchia?\" \u0159\u00edk\u00e1 f \u017eensk\u00fd hlas. \"Remember Pinocchio?\" says f a female voice." }, "FIGREF1": { "num": null, "type_str": "figure", "uris": null, "text": "\"Porovnejte tyto dva kandid\u00e1ty na starostu,\".Effect f \u0159\u00edk\u00e1 hlasatel. \"Compare two candidates for mayor,\".Effect f says the announcer." }, "FIGREF2": { "num": null, "type_str": "figure", "uris": null, "text": "The war over federal judicial salaries takes a victim.\u2193 V\u00e1lka o platy feder\u00e1ln\u00edch soudc\u016f si \u017e\u00e1d\u00e1 svou prvn\u00ed ob\u011b\u0165 f ." }, "TABREF0": { "num": null, "html": null, "content": "
[Context: Moje matka m\u00e1 r\u00e1da
r\u016f\u017ee a tulip\u00e1ny.] Tulip\u00e1ny contrastive_topic matka topic v\u010dera topic koupila focus na tr
hu focus
Literally: [Context: My mother likes
roses and tulips.] The tulips contrastive_topic the mother topic yes
terday topic
", "text": "The mother bought the tulips ON THE MARKET 3 yesterday.)", "type_str": "table" }, "TABREF8": { "num": null, "html": null, "content": "", "text": "Usage of the individual pre-annotation steps", "type_str": "table" }, "TABREF9": { "num": null, "html": null, "content": "
", "text": "The distribution of changes of automatically pre-annotated TFA-values manually made by human annotators", "type_str": "table" } } } }