Documentation to CzeDLex
CzeDLex is a lexicon of Czech discourse connectives, originally (2015 – 2017) developed within the COST-cz project TextLink-cz (see Czech project notes for individual years: 2015, 2016, 2017).
Its pilot version –
CzeDLex 0.5 – was published in December 24, 2017 in the Lindat/Clarin repository. It is also available on-line.
The lexicon contains connectives partially automatically extracted from the Prague Discourse Treebank 2.0 (PDiT 2.0), a large corpus annotated manually with discourse relations. The lexicon entries are being manually checked and supplemented by additional information and English translations, starting with the most frequent connectives. The current development version of the lexicon is available on-line.
License and Availability
CzeDLex is publicly available under the Creative Commons - Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.
CzeDLex is available in two formats, PML and HTML:
The Prague Markup Language (PML) is the primary XML format of the lexicon. The lexicon (version 0.5) in this format is dowloadable from the Lindat/Clarin repository and can be opened (browsed and edited) in the tree editor TrEd. Installation instructions are a part of the Lindat/Clarin package.
The on-line version of the lexicon in the form of HTML web pages presents the most important properties of the lexicon entries in a graphical, user-friendly way, without a need to install any tools.
The on-line version of the lexicon allows to filter the list of lexicon entries by three criteria (which cannot be combined): the basic filter distinguishes the primary and secondary connectives, the second filter distinguishes the connectives according to discourse types they are able to express, and the last filter distinguishes the connectives according to their part of speech.
How to cite
If you use the data of the lexicon or wish to refer to the published version of the data (version 0.5), please cite the publication of the data:
Mírovský Jiří, Synková Pavlína, Rysová Magdaléna, Poláková Lucie: CzeDLex 0.5. Data/software, Charles University, Prague, Czech Republic, http://hdl.handle.net/11234/1-2538, Dec 2017
You can also cite the following journal article describing the design of the lexicon and the extraction of the lexicon from the source corpus:
Mírovský Jiří, Synková Pavlína, Rysová Magdaléna, Poláková Lucie: CzeDLex – A Lexicon of Czech Discourse Connectives. In: The Prague Bulletin of Mathematical Linguistics, No. 109, Univerzita Karlova v Praze, Prague, Czech Republic, ISSN 0032-6585, pp. 61-91, Oct 2017
Manual Checks and Additions
Manual checks included checking of auto-filled values, assessment of suspicious usages of the connectives (in terms of complex forms – see below in the description of complex forms at level-two entries), addition of attributes/elementes not filled in automatically, translation of level-one entries to English, addition of glosses to individual usages (and their translations), translation of complex forms and modifications, selection of their types, selection of the most appropriate examples and their translation to English. Substantial information that could not be added within structural attributes/elements was provided as a free text in the element note.
In total, there are 205 level-one entries in the lexicon. The following lexicon entries (covering more than 2/3 of the discourse relations annotated in the PDiT 2.0) have been fully manually checked and supplemented with additional information in CzeDLex 0.5 (in brackets: numbers of connective usages in the PDiT 2.0, incl. variants, modifications and complex forms): a [and] (6 617), ale [but] (1 746), však [however] (1 686), když [when] (780), protože [because] (637), totiž [actually, you see, you know] (485), proto [therefore] (478), pokud [if] (473), aby [(in order) to] (451), pak [then] (432), : (416), tedy [thus, so, namely, I mean] (339), také [also, too] (334), tak [so, in that way] (331) and ovšem [but, of course] (311), and secondary connectives with core words dodat [add] (187), případ [case] (83) and vzhledem k [with respect to] (42). Realizations at several more secondary connectives have been manually sorted according to a dependency scheme: důvod [reason] (76), strana [side] (61), naproti [on the other hand] (24), oproti [on the other hand] (4).
The level-one entry in the lexicon structure is represented by the lemma of the connective. It is encoded in the element lemma and contains the following information:
- element text: the lemma of the connective
- element english: an approximate English translation for a basic orientation; more precise translations are given in connection with semantic discourse types at level-two entries
- element type: the type of the connective: primary vs. secondary
- element struct: the structure of the connective: it signals whether the connective is single such as proto [therefore] or complex such as jednak jednak [on the one hand on the other hand]. The complex connectives are further differentiated in the attribute type according to their placement in the argument(s): complex connectives with parts occurring in both arguments (e.g. jednak jednak [on the one hand on the other hand] or buď nebo [either or]) are labeled correlative, while complex connectives with all parts occurring in a single argument are labeled continuous if no word can be inserted between the parts of the connective (e.g. the connective i když [even if, although]), or discontinuous if other words can occur between the connective parts (e.g. a potom [and then]). Multiplied connectives in coordinations (e.g. protože ... and protože [because ... and because]) are labeled as multiple.
- element variants: a list of variants of the connective: they are further specified in the attribute type as stylistic (cf. neutral tedy [so.neutral] vs. informal teda [so.informal]) or orthographic (e.g. mimoto vs. mimo to [both meaning: besides]), or inflection (e.g. the form čímž [by which] is the instrumental form of the connective with the nominative form což [which])
- element conn-usages: a list of connective usages – level-two entries
- element non-conn-usages: a list of non-connective usages – level-two entries
- element note: important information not encoded in other attributes
- attribute id: a lexicon-wide unique identifier of this level-one lexicon entry
- element src: an identifier of an annotator editing this lexicon entry
- element is_checked: is set to 1 for entries considered to be fully checked and annotated
For each level-one entry in the lexicon structure, its connective and non-connective usages are represented as level-two entries. In connective-usages, the discourse type (see Table 1) is used as the base for nesting, while in non-connective-usages, the part-of-speech appurtenance of the expressions is used. The second level entry of the lexicon is encoded in the element usage and contains the following information:
- element sense: the discourse type (see Table 1)
- element scheme: the dependency scheme (used for secondary connectives only)
- element gloss: a Czech expression disambiguating the meaning of the connective (a synonym or an explanatory phrase)
- element english: an English translation (the gloss in English)
- element pos: the part-of-speech appurtenance of the connective (the lemma) in the given usage. Conjunctions are further distinguished in the attribute subpos as coordinating or subordinating.
- element syntax: for secondary connectives, the part-of-speech characteristics of the core word is accompanied by a syntactic characteristics for the whole secondary connective represented by this usage (nominal phrase, adjectival phrase, pronominal phrase, clause, adverbial phrase, or prepositional phrase).
- element arg_semantics: this characteristics specifies the semantics of the argument the connective occurs in (see Table 2). From the semantic perspective, there is a basic difference between symmetric and asymmetric discourse relations. While both arguments of a symmetric relation (i.e. conjunction or synchrony) share the same general semantic characteristics, asymmetric discourse relations (e.g. reason–result or gradation) hold between arguments that have different semantic nature (e.g. one argument expresses the reason, the other the result). A connective of an asymmetric relation is characterized by its placement in one specific part of the relation it signals. For example, the coordinating conjunction tedy [thus] signals the result, while totiž [because] signals the reason. Similarly, the subordinating conjunctions než [until] and když [when] can be used for signalling precedence–succession – the former occurs in the argument expressing the event happening later, while the latter occurs in the argument expressing the earlier event. For symmetric relations, the element arg_semantics has the value symmetric. For complex correlative connectives forming level-one entries, the value is given for the second part of the connective.
- element ordering: signals the linear order of the argument the connective occurs in (relatively to the other – external – argument). In the majority of cases, ordering is connected with the part-of-speech characteristics – coordinating conjunctions, adverbs and particles are placed in the second argument in the linear order, while subordinating conjunctions can be placed in either of the arguments. There are, however, exceptions – e.g. the particle nejenže [not only that] which occurs always in the first argument – that justify incorporation of this characteristics as a separate element into the lexicon. The element ordering has one of these five values: 1 for connectives occurring only in the first argument, 2 for connectives in the second argument, 1 or 2 for connectives in the first or second argument, 1 and 2 for complex correlative connectives and N/A for secondary connectives forming a separate syntactic unit (e.g. Důvod je jednoduchý. [The reason is simple.]) and therefore occurring entirely between the arguments.
- element integration: captures the position of the connective within the argument. According to their origin and other possible functions in text, Czech connectives have different positions in the argument. Only subordinating conjunctions and prototypical coordinating conjunctions occupy the very beginning of the clause or sentence; the position of other connectives varies. Some of them are placed typically at the clitic, i.e. second position (e.g. však [however]), some of them are typically either on the first or on the second position (e.g. potom [then] or proto [therefore]) and for the class of focusing particles (i.e. expressions like také [also] or jenom [only]), the position is given by the information structure. For secondary connectives represented by the whole clause, integration is again N/A. Other values of this element, as follows from examples just mentioned, are first, second, first or second, and any. For complex correlative connectives forming level-one entries, the value is given for the second part of the connective only.
- element realizations: a list of non-modified and non-complex secondary connectives from PDiT 2.0 represented by the given dependency scheme (applies only to secondary connectives)
- element modifications: a list of the connective modifications: e.g. for the lemma potom [then] expressing precedence–succession, there is a modification teprve potom [only then]. Secondary connectives can be modified as well – cf. hlavní důvod proč [the main reason why]. Modifications are further distinguished in the attribute type as eval (evaluative), modal, and intense (intensifying).
- element complex_forms: a list of complex connectives: e.g. for the lemma potom [then] expressing precedence–succession, there are for example complex forms a potom [and then] and nejdřív potom [first then]. Secondary connectives can have complex forms as well – cf. a z tohoto důvodu [and for this reason]. The criterion for a complex form to be placed in the level-two entry under a certain lemma is the ability of the basic connective (the given lemma) to express the same discourse type. It means that e.g. the complex connective přesto však [yet however] expressing the discourse type of concession is placed in respective level-two entries under both lemmas přesto [yet] and však [however], because both these single connectives individually also express the discourse type of concession in PDiT 2.0. Further, according to its placement either in both arguments or in one argument, each complex form is labeled in the attribute type as correlative, continuous, discontinuous or multiple (see above among the level-one entry characteristics). Within each complex form, element note may contain additonal information.
- element examples: a list of a few illustrative examples from PDiT 2.0 and their English translations. Both intra-sentential and inter-sentential examples are – if available in the corpus – given for the connective usages and marked as such in the attribute type (intra vs. inter).
- element is_rare: signals a rare use of the connective with the given discourse type
- element register: captures whether the connective is used in the neutral, formal or informal register
- element note: important information not encoded in other attributes
- attribute id: a unique identifier of this level-two entry
Table 1: List of possible discourse types (senses)
Table 2: Possible values of the argument semantics (attribute arg_semantics)
condition:result of condition
||pragmatic condition:pragmatic condition
pragmatic condition:result of pragmatic condition
||pragmatic reason-result:pragmatic reason
pragmatic reason-result:pragmatic result
||restrictive opposition:general statement
|all other relations
Numbers of occurrences in the PDiT 2.0 were added to all individual variants, complex forms, modifications and realizations, as well as to connective and non-connective usages (level-two entries) and the whole lemmas (level-one entries), in two attributes: pdt_count and pdt_intra, capturing numbers of all vs. intra-sentential occurrences of the respective items.
Apart from English translations listed in the descriptions of level-one and level-two entries, all complex forms, modified forms, realizations, variants (when possible) and examples have been translated to English (the translations are captured in elements english at the respective places).