| { | |
| "paper_id": "2021", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T07:42:50.566490Z" | |
| }, | |
| "title": "A morph-based and a word-based treebank for Beja", | |
| "authors": [ | |
| { | |
| "first": "Sylvain", | |
| "middle": [], | |
| "last": "Kahane", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "INALCO", | |
| "location": { | |
| "region": "& EPHE" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Martine", | |
| "middle": [], | |
| "last": "Vanhove", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Rayan", | |
| "middle": [], | |
| "last": "Ziane", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Bruno", | |
| "middle": [], | |
| "last": "Guillaume", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "The paper presents the first UD treebank for Beja, a Cushitic language spoken in Sudan. It has been built from the conversion and enhancement of an Interlinear Glossed Text (IGT). The paper's objectives are threefold: we explain our choice to use a morph-based annotation and its consequences, we describe the processing chain from an IGT to a morph-based dependency treebank and a word-based treebank, and we present several interesting constructions in Beja.", | |
| "pdf_parse": { | |
| "paper_id": "2021", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "The paper presents the first UD treebank for Beja, a Cushitic language spoken in Sudan. It has been built from the conversion and enhancement of an Interlinear Glossed Text (IGT). The paper's objectives are threefold: we explain our choice to use a morph-based annotation and its consequences, we describe the processing chain from an IGT to a morph-based dependency treebank and a word-based treebank, and we present several interesting constructions in Beja.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "This paper presents a small treebank for Beja, a Cushitic language spoken in Sudan. Initially developed in SUD (Surface-Syntactic Universal Dependencies) (Gerdes et al. 2018 (Gerdes et al. , 2019 , the treebank is also available in UD (de Marneffe et al. 2021) . It has been built from an Interlinear Glossed Text (IGT) (Comrie et al. 2008) developed by Martine Vanhove (2014) . The original corpus contains 5899 words and 12507 morphs representing slightly less than one hour of recordings divided into 18 files. Two files, containing 1101 morphs, 418 stems, and 56 sentences, have been completely annotated and constitute the UD2.8 Beja-NSC treebank. 1 One of our goals was to avoid losing information contained in the original resource, which led us to adopt a morph-based tokenization rather than a word-based annotation.", | |
| "cite_spans": [ | |
| { | |
| "start": 154, | |
| "end": 173, | |
| "text": "(Gerdes et al. 2018", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 174, | |
| "end": 195, | |
| "text": "(Gerdes et al. , 2019", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 235, | |
| "end": 260, | |
| "text": "(de Marneffe et al. 2021)", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 320, | |
| "end": 340, | |
| "text": "(Comrie et al. 2008)", | |
| "ref_id": "BIBREF2" | |
| }, | |
| { | |
| "start": 362, | |
| "end": 376, | |
| "text": "Vanhove (2014)", | |
| "ref_id": "BIBREF21" | |
| }, | |
| { | |
| "start": 653, | |
| "end": 654, | |
| "text": "1", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The paper focuses on three aspects of developing this treebank. Section 2 presents the Beja corpus and the IGT annotation we started with, the UD annotation scheme, and the adjustments to it which were necessary for our morph-based annotation. An overview of the conversion to a word-based treebank is also provided. Section 3 explains the processing chain from an IGT to a UD treebank and the optimization of this chain. Section 4 introduces some challenges faced during the syntactic annotation of Beja.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Beja is the sole member of the North Cushitic branch of the Afroasiatic phylum. It is mostly spoken in eastern Sudan, as well as in southern Egypt and northern Eritrea. In Sudan, the country where the data were collected, the number of speakers is about 2,000,000, but the language has no official recognition and exists purely as an oral language. As explained by Martine Vanhove (2006) , Beja is not poorly described compared to other Sudanese languages, and the most recent grammar, published in French, goes back to 2017 (Vanhove 2017) . However, some elements required for a complete description of the language remain unavailable.", | |
| "cite_spans": [ | |
| { | |
| "start": 373, | |
| "end": 387, | |
| "text": "Vanhove (2006)", | |
| "ref_id": "BIBREF20" | |
| }, | |
| { | |
| "start": 525, | |
| "end": 539, | |
| "text": "(Vanhove 2017)", | |
| "ref_id": "BIBREF22" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Beja and CorpAfroAs", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "The data used for the development of this treebank comes from the CorpAfroAs project (Mettouchi & Chanard 2010) . CorpAfroAs is a multilingual corpus which aimed at providing a structured database of natural records of Afroasiatic languages, transcribed, translated and annotated to allow for complex requests. CorpAfroAs is organized around two axes: prosodic analysis and morphosyntactic glossing. It is this morphosyntactic glossing that served as the raw material for our work.", | |
| "cite_spans": [ | |
| { | |
| "start": 85, | |
| "end": 111, | |
| "text": "(Mettouchi & Chanard 2010)", | |
| "ref_id": "BIBREF16" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Beja and CorpAfroAs", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "All CorpAfroAs corpora use a common format for IGTs presented in Comrie (2015) .", | |
| "cite_spans": [ | |
| { | |
| "start": 65, | |
| "end": 78, | |
| "text": "Comrie (2015)", | |
| "ref_id": "BIBREF1" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Beja's Interlinear Glossed Text", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "(1) w=\u0294i\u02d0d arraf-i / a-di=t a-ba i-ni // DEF=Aid congratulate-AOR.1SG 1SG-say\\PFV=COORD 1SG-go 3SG.M-say\\PFV 'I went to wish him a blessed Aid, he said.' Figure 1 . ELAN screenshot of the CorpAfroAs IGT for 1The top of Figure 1 contains the timeline. Tier 1 is the segmentation into prosodic units. Tier 2 is a broad phonetic transcription; / marks a minor prosodic break (rising final continuative intonation) and // a major prosodic break (falling final intonation). Tier 3 contains the prosodic words and Tier 4 a tokenization into the smaller units with a non-segmentable signifier (i.e. the morphs, as defined by Haspelmath 2020). The tokens considered are lexical stems, inflected forms of stems (when the inflectional \"morphemes\" are not affixes), inflectional affixes, and derivational affixes. For the sake of simplicity, we will call these tokens morphs, even if some of them are a combination of morphemes, such as ni, which is the perfective form of the stem of the verb di 'to say'. This inflected stem also combines with a prefix i-, which is a subject agreement morph, giving the verbal form ini at the end of example (1).", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 154, | |
| "end": 162, | |
| "text": "Figure 1", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 219, | |
| "end": 227, | |
| "text": "Figure 1", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Beja's Interlinear Glossed Text", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "Tier 5, labelled GE, is a gloss. Tier 6, labelled RX, contains morphosyntactic features, including POS (DET, N, V1\u2026) and inflectional categories (TAM, PNG for Person-Number-Gender, \u2026).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Beja's Interlinear Glossed Text", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "Tier 7 is a translation of each prosodic unit and Tier 8, called the \"major free translation\" (MFT), is a translation based on larger units, allowing for better translations. It is this last tier which was used as the basis for the sentence segmentation. Since the end of each MFT unit does not necessarily corresponds to the end of a sentence, understood as a coherent syntactico-semantic unit, we copied the MFT tier (Mft-cp) on which we signaled the end of each sentence by a # sign ( Figure 2 ). 2", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 488, | |
| "end": 496, | |
| "text": "Figure 2", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Beja's Interlinear Glossed Text", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "The IGT does not contain lemmas (which supposes a more advanced description and a lexicon), but contains glosses and translations. The corpus contains a time alignment and each IGT sample is coupled with a sound file accessible on the CorpAfroAs website.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Figure 2. ELAN screenshot of the sentence segmentation", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "T1 T2 T3 (word) T4 (morph) T5 (GE) T6 (RX) T7 T8 (MFT)", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Figure 2. ELAN screenshot of the sentence segmentation", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "We use the \"morph\" segmentation for tokenization. The content of the tiers GE and RX is kept in features GE and RX. The time alignment gives us the features AlignBegin and AlignEnd of each token, including the prosodic breaks ( Figure 3 ) (see Kahane et al. 2021 for the conventions we use for spoken corpora).", | |
| "cite_spans": [ | |
| { | |
| "start": 244, | |
| "end": 262, | |
| "text": "Kahane et al. 2021", | |
| "ref_id": "BIBREF14" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 228, | |
| "end": 236, | |
| "text": "Figure 3", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "A morph-based annotation scheme for SUD and UD", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "'I said and I went, he said' Figure 3 . The UD annotation for the end of (1) Some pieces of information of the GE tier are used to instantiate morphosyntactic features, such as Number, Person, Gender, Aspect, Definiteness\u2026 (see Section 3.2). A TokenType is added on each token, with five values: Stem for stems (including some inflected stems), Clit for clitics, InflAff for inflectional affixes, DerAff for derivational affixes, and Break for prosodic breaks.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 29, | |
| "end": 37, | |
| "text": "Figure 3", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "A morph-based annotation scheme for SUD and UD", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "In addition to the usual #text and #sent_id features (see online UD guidelines or de Marneffe et al. 2021), the metadata contain a #sound_url feature for the URL of the sound file corresponding to the transcription and #phonetic_text for the phonetic transcription from Tier 2 (Kahane et al. 2021) . The #text value, which is the concatenation of the tokens, including simple and double hyphens, is distinct from the #phonetic_text value.", | |
| "cite_spans": [ | |
| { | |
| "start": 277, | |
| "end": 297, | |
| "text": "(Kahane et al. 2021)", | |
| "ref_id": "BIBREF14" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "A morph-based annotation scheme for SUD and UD", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "We decided to give each token a POS as if it was a word. In consequence, pronominal affixes are PRON, verb nominalizers are SCONJ, TAM or causatives are AUX, case markers are ADP, plurals are DET, and purely phonetic signs are PART.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "A morph-based annotation scheme for SUD and UD", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "In our SUD annotation, SCONJ, ADP, and AUX affixes are treated as governors of their base, as if they were words (Gerdes et al. 2018) . We use the corresponding SUD syntactic relation with the aff extension: In other words, subject pronominal affixes are subj:aff of their base, plurals are det:aff, while for SCONJ, ADP, or AUX, the base is comp:aff of the affix. Figure 4 gives the example of the deverbal noun siganfo:j 'settling' where the verbal stem ganf 'make kneel' combines with two derivational affixes, the causative prefix si-and the nominalizer -o:j, 'settling' Note that the fact that some affixes are treated as heads allows us to indicate in which order they combine. In the case of siganfo:j, ganf combine first with the causative si-and then with the nominalizer -o:j. Derivational affixes receive an additional feature DerPos indicating the POS of the derived form: AUX affixes have a DerPos=VERB and SCONJ affixes have a DerPos=NOUN. For inflected forms, the POS of the inflected form remains the POS of the stem. Note also that in the case of an inflected form, the base can have its own dependents, while in the case of a derived form all dependents are on the derivational affix. 3", | |
| "cite_spans": [ | |
| { | |
| "start": 113, | |
| "end": 133, | |
| "text": "(Gerdes et al. 2018)", | |
| "ref_id": "BIBREF6" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 365, | |
| "end": 373, | |
| "text": "Figure 4", | |
| "ref_id": "FIGREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "A morph-based annotation scheme for SUD and UD", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "s da ji upos=ADJ AlignBegin=85339 AlignEnd=85497 GE=good Gloss=good RX=[ADJ] TokenType=Stem =t upos=DET AlignBegin=85497 AlignEnd=85656 Definite=Ind GE==[INDF].[F] Gender=Fem RX==[DET] TokenType=Clit si- upos=AUX AlignBegin=85656 AlignEnd=85761 DerPos=VERB GE=[CAUS]- RX=[V1].[DER]- TokenType=DerAff VerbClass=1 Voice=Cau ganf upos=VERB AlignBegin=85761 AlignEnd=85867 GE=make_kneel Gloss=make_kneel RX=[V1] TokenType=Stem VerbClass=1 -o j upos=SCONJ AlignBegin=85867 AlignEnd=85973 DerPos=NOUN GE=-[N].[AC] RX=-[N].[V", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "A morph-based annotation scheme for SUD and UD", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "In the UD version of the morph-based, all affixes are dependent of the stem. We use the UD syntactic relation corresponding to their functional role, with an additional aff extension: subject pronominal affixes are nsubj:aff, case markers are case:aff, nominalizers are mark:aff, AUX affixes are aux:aff, plurals are det:aff. Figure 5 gives the UD version of the two words of Figure 4 . (See Figure 3 of the example of pronominal affixes in UD-style.) Note that the order in which the affixes combine with the stem is lost in the UD version. This is a problem for the conversion to the word-based version, because we cannot easily determine whether siganfo:j is a noun or a verb (see for a similar discussion about the fact that UD underspecifies the internal structures of nuclei). For this reason, the word-based UD version of the treebank is derived from the morph-based and word-based SUD versions and not from the morph-based UD version. ", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 326, | |
| "end": 334, | |
| "text": "Figure 5", | |
| "ref_id": "FIGREF1" | |
| }, | |
| { | |
| "start": 376, | |
| "end": 384, | |
| "text": "Figure 4", | |
| "ref_id": "FIGREF0" | |
| }, | |
| { | |
| "start": 392, | |
| "end": 400, | |
| "text": "Figure 3", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "A morph-based annotation scheme for SUD and UD", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "UD theoretically requires treebanks to be word-based. However, we think that the fact that our treebank is morph-based does not pose a problem because the morph-based annotation is explicit, due to the different features we introduced (TokenType on tokens and aff on relations) and because it is not difficult to merge every stem with its affixes to obtain a word-based treebank. Note that the question to have morph-based (generally called morpheme-based) treebank rather than word-based treebank has been discussed several times for different languages: see Tsarfaty & Goldberg (2008) for Modern Hebrew, Vincze et al. (2010) for Hungarian, Zhan et al. (2014) for Chinese, or Park (2017) for Korean. 4 For the conversion into a word-based treebank, the lists of morphosyntactic features attached to the different parts of a word must be merged in different ways. Some features had to be concatenated, such as the feature form, containing the form, and the features GE and RX containing the morphosyntactic glosses. See Figure 6 for the word-based version of Figures 4 and 5. His cleaning of the table was impressive. In (i), cleaning is an inflected form and clean can have a subject and a direct object, while, in (ii), cleaning is a derived noun with a determiner and a noun complement. 4 We would like to thank one of our reviewers, who pointed out that our segmentation was morph-based rather than morphemebased. The most challenging feature is upos (the UD feature for the \"universal\" POS) for derived forms. Thanks to the SUD annotation where the derivational affixes are head and to the DerPos feature, it becomes trivial to compute the upos of a derived form: it is the DerPos of the topmost derivational affix. Except for the features form, GE, and RX, which are concatenated, and upos, which is replaced by DerPos, the features of the derived form are the features of the topmost derivational affix.", | |
| "cite_spans": [ | |
| { | |
| "start": 560, | |
| "end": 586, | |
| "text": "Tsarfaty & Goldberg (2008)", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 606, | |
| "end": 626, | |
| "text": "Vincze et al. (2010)", | |
| "ref_id": "BIBREF23" | |
| }, | |
| { | |
| "start": 701, | |
| "end": 702, | |
| "text": "4", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 1290, | |
| "end": 1291, | |
| "text": "4", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 1020, | |
| "end": 1028, | |
| "text": "Figure 6", | |
| "ref_id": "FIGREF2" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "From a morph-based treebank to a word-based treebank and back", | |
| "sec_num": "2.4" | |
| }, | |
| { | |
| "text": "For inflected forms, the upos of the word is the upos of the stem. The features form, GE, and RX are concatenated as for derived forms, but contrary to derived forms, other features are unified for inflected forms. Figure 7 shows the word-based version of the morph-based analysis of Figure 3 , where we can see that the Person and Number features coming from the pronominal affixes are reported on the verb forms. Figure 7 . The word-based UD annotation for the end of (1)", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 215, | |
| "end": 223, | |
| "text": "Figure 7", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 284, | |
| "end": 292, | |
| "text": "Figure 3", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 415, | |
| "end": 423, | |
| "text": "Figure 7", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "From a morph-based treebank to a word-based treebank and back", | |
| "sec_num": "2.4" | |
| }, | |
| { | |
| "text": "We faced one unexpected problem with a clitic placed between a stem and an inflectional affix. We analyzed this case as an amalgam with only one word corresponding to two lexemes (Figure 8 ).", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 179, | |
| "end": 188, | |
| "text": "(Figure 8", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "From a morph-based treebank to a word-based treebank and back", | |
| "sec_num": "2.4" | |
| }, | |
| { | |
| "text": "'our grand-father' 'our grand-father' Figure 8 . The morph-based and word-based SUD annotation of an incorporated clitic", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 38, | |
| "end": 46, | |
| "text": "Figure 8", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "From a morph-based treebank to a word-based treebank and back", | |
| "sec_num": "2.4" | |
| }, | |
| { | |
| "text": "We must also compute the AlignBegin and AlignEnd features of words, which are easily deductible from the corresponding features of the morphs. Note that in the word-based version of the treebank, some information is lost and it will not be possible to recover the segmentation into morphs, as well as the features associated to this morphs. This is why we decided to distribute the Beja [email protected] treebank in a morph-based version. On our side, we will maintain the morph-based SUD version, which is the most informative one (see in particular the discussion of Section 3 when the word contains two derivational affixes).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "From a morph-based treebank to a word-based treebank and back", | |
| "sec_num": "2.4" | |
| }, | |
| { | |
| "text": "The construction of the UD treebank was carried out in three steps: the conversion of the IGT into a CoNLL-U (Section 3.1); the automatic pre-annotation by enrichment of the CoNNL-U (Section 3.2); the manual SUD annotation, the conversion to UD and the validation of the treebank (Section 3.3). ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "From IGT to UD", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "The first obstacle to the conversion to CoNLL-U is the fact that the corpus is segmented into prosodic units that do not necessarily correspond to syntactic units. We decided to base our major segmentation on the \"major free translation\" segmentation, which corresponds more or less to illocutionary units as illustrated in Figure 2 in section 2.2. The tokenisation is based on the segmentation into \"morphs\". Once the choice of the tiers for the tokenization and the segmentation into sentences is made, the conversion of the IGT to a CoNLL-U is straightforward and loses no information from the IGT format. For each token, the time alignment is stored in the features AlignBegin and AlignEnd, the content of the GE and RX tiers is stored in the features GE and RX.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 324, | |
| "end": 332, | |
| "text": "Figure 2", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "From IGT to CoNLL-U", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "The first CoNLL-U we obtain is almost similar to the IGT. The second step consists in enriching this CoNLL-U by transferring the content of the GE and RX tiers into UD features in order to fit the UD annotation scheme. The annotation specific to the morph-based level was introduced entirely automatically.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Automatic pre-annotation", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "As the GE and RX formats of CorpAfroAs IGTs are enriched versions of the Leipzig Glossing Rules (Comrie 2015), they allow us to infer the UD POS and all the UD morphosyntactic features that must be associated with the tokens. We built a lexicon that proposes a translation into a UD feature for each label used in the GE and RX tiers. It was also possible to infer the syntactic relation for many tokens. The Grew tool (Guillaume, 2021) , through its graph rewriting function, makes it possible to write a grammar of rules matching elements within dependency trees.", | |
| "cite_spans": [ | |
| { | |
| "start": 419, | |
| "end": 436, | |
| "text": "(Guillaume, 2021)", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Automatic pre-annotation", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "The feature TokenType, which distinguishes stems, affixes, clitics, and prosodic breaks, is based on the form of the token: As usual in IGTs, affixes have a hyphen (a-or -a) and clitics a double hyphen (ba= or =i), while prosodic breaks are assigned to special symbols (/ and //). For affixes and clitics, the governor was the closest stem and the positions of the hyphens indicate if the stem occurs after or before them. The distinction between inflectional and derivational affixes can be computed from the syntactic RX feature (most derivational affixes have a DER value in RX).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Automatic pre-annotation", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "The syntactic label set of CorpoAfroAs, corresponding to the RX tier, is richer than the UD upos set of POS and the POS conversion was trivial for most of the labels. For instance, the labels V1, V2, LV, and IRG are all converted to the VERB upos tag. In order not to lose information, V1 and V2 receive a VerbClass feature with values 1 and 2 according to the original label. LV is provided with a VerbType=Light feature. In a similar way, the label DEM is converted into a DET upos with the PronType=Dem feature. The label REL for relativizers gives us an SCONJ upos, as well as the SUD relation mod@relcl (translated into UD acl:relcl). Moreover, due to the head-final behaviour of Beja, the relativizer can be linked to the verb preceding it (Figure 9 ). 5 'a camel that I had let loose' Figure 9 . Relative clause (UD-style) 5 This figure, as well as all the following figures, is extracted from the morph-based UD version of the treebank, revised for the paper, and which will be distributed on May 1 st , 2022, for the UD2.10 release. All the SUD versions of the treebanks are available on the SUD website https://surfacesyntacticud.github.io/ and all versions can be requested on the Grew-match website (the latest versions of UD treebanks converted from SUD treebanks are at the end of the list of UD treebanks). ", | |
| "cite_spans": [ | |
| { | |
| "start": 830, | |
| "end": 831, | |
| "text": "5", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 746, | |
| "end": 755, | |
| "text": "(Figure 9", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 792, | |
| "end": 800, | |
| "text": "Figure 9", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Automatic pre-annotation", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "The manual annotation was carried out in the SUD annotation scheme by a linguist specializing in Beja (Martine Vanhove), with the help of a specialist of treebank annotation (Sylvain Kahane) and a master student in NLP (Rayan Ziane), as well as some feedback from two native speakers (Ahmed Mohamed-Tahir Hamid and Mohamed-Tahir Hamid Ahmed). It was not possible to have a double annotation for this language. Some problems of analysis we faced during the annotation process are discussed in the next section. For the conversion from SUD to UD, we had to customize the conversion of the relations introduced for the affixes. The fact that UD forces the coordination relation conj to be head-initial was also a problem and SUD head-final conj relations were converted into an ad hoc dep:conj UD relation (see Section 4.2). The different conversions were mastered by Rayan Ziane and Bruno Guillaume.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Manual SUD annotation and conversion from SUD to UD", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "Below, we discuss four features of Beja syntax: affixes and word order (4.1), coordination (4.2), relative clauses (4.3) and serial verb constructions (4.4).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Some constructions of Beja", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The Beja treebank contains 684 words if we count both stems and clitics; 39% of words are clitics. The treebank contains 244 affixes for 418 stems, or a proportion of 58%. 59% of them are suffixes and 41% prefixes. 88% of the affixes are on verbs, 7% on nouns, and 5% on auxiliaries. All affixes on nouns are suffixes. Not all inflectional morphemes are affixes: 44% of the stems are in fact inflected forms, containing an inseparable inflectional morpheme, which increases the proportion of inflectional morphemes to 102% (102 inflectional morphemes for 100 stems).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Affix and word order", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "Beja is a head-final language: only 11% of the dependencies between two stems have the governor before the dependent in the SUD version of the treebank. Among the 31 dependencies concerned, 11 are for modifiers, 6 for discourse markers, 4 for dislocated objects, 2 for objects in canonical position, and 2 for determiners. Clitics occur on both sides: 47% are proclitics and 53% are enclitics. Clitics are mainly on verbs (56%) and nouns (38%). Clitics on nouns are determiners (70%), possessives (15%), postpositions (11%), and coordinating conjunctions (3%). Clitics on verbs are subordinating conjunctions (35%), object pronouns (25%), coordinating conjunctions (14%), an optative particle (2%), and, on nominalized forms, determiners (15%) and copulas (8%).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Affix and word order", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "Beja adpositions are postpositions, either as independent words ( Figure 10 ) or as enclitics ( Figure 11 ", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 66, | |
| "end": 75, | |
| "text": "Figure 10", | |
| "ref_id": "FIGREF4" | |
| }, | |
| { | |
| "start": 96, | |
| "end": 105, | |
| "text": "Figure 11", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Affix and word order", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "In Beja, verbal and nominal coordination are expressed with different enclitic morphemes. The texts contain 20 tokens of the verbal coordinating conjunction =t (and its allomorphs =it and =ajt) ( Figure 13 ). For half of the tokens, the conjunctions occur at the end of a prosodic unit, be it a major or a minor prosodic break, or a sentence. For this reason, we attach the coordinating conjunctions to the first conjunct and we consider that the second conjunct is the head of the coordination. See Kanayama et al. (2018) for a similar analysis in two other head-final languages, Japanese and Korean. As the conj relation is forbidden from right to left in UD, we introduced a dep:conj relation.", | |
| "cite_spans": [ | |
| { | |
| "start": 501, | |
| "end": 523, | |
| "text": "Kanayama et al. (2018)", | |
| "ref_id": "BIBREF15" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 196, | |
| "end": 206, | |
| "text": "Figure 13", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Coordination in a head-final language", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "'they took them and they came to me' Figure 13 . Verbal coordination (UD-style)", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 37, | |
| "end": 46, | |
| "text": "Figure 13", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Coordination in a head-final language", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "The verbal coordinating conjunction is tightly linked to the right of the verb. It even occurs before enclitic object pronouns (6 tokens), as in Figure 14 .", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 145, | |
| "end": 154, | |
| "text": "Figure 14", | |
| "ref_id": "FIGREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Coordination in a head-final language", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "'you told me and \u2026' Figure 14 . Position of the verbal coordinating conjunction (UD-style)", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 20, | |
| "end": 29, | |
| "text": "Figure 14", | |
| "ref_id": "FIGREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Coordination in a head-final language", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "The nominal coordinating conjunction is the enclitic wa. It is expressed on each conjunct as shown in Figure 15. 'a sword and the stick' Figure 15 . Nominal coordination (UD-style)", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 102, | |
| "end": 112, | |
| "text": "Figure 15.", | |
| "ref_id": "FIGREF1" | |
| }, | |
| { | |
| "start": 137, | |
| "end": 146, | |
| "text": "Figure 15", | |
| "ref_id": "FIGREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Coordination in a head-final language", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "In the texts, only object relative clauses occur, for which the number of tokens amounts to 18. They are marked by several clitics, either proclitics (ji=, j=, wi=, w=), or enclitics (=e\u02d0b, =e\u02d0t, =t, =b, =e\u02d0) . There are also instances of a zero morph. 7 tokens were found with a preposed antecedent and 11 tokens with a postposed antecedent.", | |
| "cite_spans": [ | |
| { | |
| "start": 183, | |
| "end": 208, | |
| "text": "(=e\u02d0b, =e\u02d0t, =t, =b, =e\u02d0)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Relative clauses", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "The anteposition of an antecedent is an unusual word order in verb-final languages. The SUD annotation revealed that this construction occurs in two contexts linked to information structuring:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Relative clauses", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "1. when the object of the transitive verb of the matrix clause is topicalized (in Figure 16 two relative clauses precede the verb of the matrix clause) 2. when the relative comes as an afterthought (Figure 17) . '(a girl) who has a fringe in front of her' Complement clauses may also be formed on the basis of a relative clause. In such cases, the antecedent, which is the dummy noun na 'thing', is always placed after the relative clause, i.e. the canonical constituent order (Figure 19 ). 'he realized that I was a man' Figure 19 . Relative-based complement clause (UD-style)", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 82, | |
| "end": 91, | |
| "text": "Figure 16", | |
| "ref_id": "FIGREF2" | |
| }, | |
| { | |
| "start": 198, | |
| "end": 209, | |
| "text": "(Figure 17)", | |
| "ref_id": "FIGREF7" | |
| }, | |
| { | |
| "start": 477, | |
| "end": 487, | |
| "text": "(Figure 19", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 522, | |
| "end": 531, | |
| "text": "Figure 19", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Relative clauses", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "We chose to label as SVCs any series of two finite verbs of the same semantic domain which are not coordinated and do not have a predicate-argument relation. This characterization only partly complies with e.g. Haspelmath's (2016: 296) narrow definition of serial verb constructions as a comparative concept: \"A serial verb construction is a monoclausal construction consisting of multiple independent verbs with no element linking them and with no predicate-argument relation between the verbs.\" Such a definition does not impose any semantic restriction on the semantic domains of SVCs (apart from expressing a dynamic event, Cleary-Kemp 2015: \u00a74.2.1.3; Haspelmath 2016: 302), as is the case in Beja. Moreover, SVCs seem to be unproductive, limited to very few semantic domains, and restricted to series of two verbs. There are 3 occurrences of SVCs in our data, 1 with a verb of saying ( Figure 20 ", | |
| "cite_spans": [ | |
| { | |
| "start": 211, | |
| "end": 235, | |
| "text": "Haspelmath's (2016: 296)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 891, | |
| "end": 900, | |
| "text": "Figure 20", | |
| "ref_id": "FIGREF9" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Non canonical Serial Verb Constructions", | |
| "sec_num": "4.4" | |
| }, | |
| { | |
| "text": "Beja belongs to a sub-family of Afroasiatic languages, the Cushitic languages, for which there were no treebanks. It is a language with a rich morphology and which, unlike its cousins, the Semitic languages, is a head-final language with non-canonical serial verb constructions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The Beja treebank that we present is a very small treebank, but richly annotated, with a segmentation into morphs, glosses, and an alignment to the sound file.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "While developing a morph-based treebank for Beja we have been led to bring some enrichments to the SUD and UD annotation schemes. We introduced the TokenType feature which takes five possible values (Stem, InflAff, DerAff, Clitic, Break) and an aff extension for syntactic relations to indicate more explicitly the internal relations of words. We also introduced the feature DerPos on derivational morphs for indicating the POS of the derived form.", | |
| "cite_spans": [ | |
| { | |
| "start": 199, | |
| "end": 237, | |
| "text": "(Stem, InflAff, DerAff, Clitic, Break)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "We have also seen that the SUD version of the morph-based treebank makes it easier to compute the word-based version of the treebank, since it explicitly indicates the internal structure of the word and the order in which the affixes combine with the stem.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "As it is possible to convert the morph-based annotation into a word-based annotation, we think it is better to distribute the morph-based annotation, which contains more information and is closer to the format that field linguists use. This format allows us to extract qualitative and quantitative information about the inflectional morphology of the language, which is extremely useful for typological studies (Greenberg 1960) .", | |
| "cite_spans": [ | |
| { | |
| "start": 411, | |
| "end": 427, | |
| "text": "(Greenberg 1960)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The Universal Dependency project was initially developed to unify treebank annotation schemes in order to have a common format for the development of NLP tools. The UD annotation scheme is heavily based on the output format developed for the Stanford parser for English (de Marneffe et al. 2006) . The 33 languages of the UD1.2 (Nivre et al. 2016) were all languages with long-standing writing traditions, and all corpora were written corpora following well-established orthographic conventions, most of them with a segmentation into words.", | |
| "cite_spans": [ | |
| { | |
| "start": 270, | |
| "end": 295, | |
| "text": "(de Marneffe et al. 2006)", | |
| "ref_id": "BIBREF4" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "UD is now integrating a wide range of new languages coming from different families. Many field linguists having data that are already analyzed in IGT are ready to enrich their corpus with a syntactic annotation. It is necessary that UD offer the possibility of a morpheme-based view of annotation, which allows them to keep the IGT structure. This paper is a first step in this direction by setting up a processing chain to convert an IGT into a morph-based treebank, then a word-based treebank.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The first version of the treebank published on May 1 st , 2021, for UD2.8, and released on November 1 st , 2021, for UD2.9, is a morph-based treebank. New modifications have been done for the publication of this article that will be incorporated for UD2.10 on May 1 st , 2022. We also plan to publish the word-based version of the treebank on the same occasion.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The segmentation of spoken corpora into major syntactic units (often called sentences, even if the notion can be problematic for spoken production) is a complex question that will not be addressed here. SeeKahane et al. (2021) for some guidelines andPietrandrea et al. (2014) for a more comprehensive study.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "We would like to thank our reviewers for their valuable remarks. The master internship of Rayan Ziane has been funded by the ANR project NaijaSynCor (2017-2021), directed by Bernard Caron.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Acknowledgements", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Serial Verb Constructions Revisited: A Case Study from Koro", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Cleary-Kemp", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Cleary-Kemp, J. 2015. Serial Verb Constructions Revisited: A Case Study from Koro. PhD dissertation, University of California at Berkeley.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "From the Leipzig Glossing Rules to the GE and RX lines", | |
| "authors": [ | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Comrie", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Corpus-based Studies of Lesser-described Languages. John Benjamins", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "207--219", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Comrie, B. 2015. From the Leipzig Glossing Rules to the GE and RX lines. In A. Mettouchi, M. Vanhove & D. Caubet (eds.), Corpus-based Studies of Lesser-described Languages. John Benjamins, 207-219.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "The Leipzig Glossing Rules: Conventions for interlinear morpheme-by-morpheme glosses", | |
| "authors": [ | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Comrie", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Haspelmath", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Bickel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Comrie, B., Haspelmath, M., & Bickel, B. 2008. The Leipzig Glossing Rules: Conventions for interlinear morpheme-by-morpheme glosses. Department of Linguistics of the Max Planck Institute for Evolutionary Anthropology & the Department of Linguistics of the University of Leipzig. Retrieved January, 28, 2010. https://www.eva.mpg.de/lingua/resources/glossing-rules.php.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "The CorpAfroAs Corpus of Spoken AfroAsiatic Languages", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Corpafroas", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1075/scl.68.website" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "CorpAfroAs: The CorpAfroAs Corpus of Spoken AfroAsiatic Languages. http://dx.doi.org/10.1075/scl.68.website.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Generating typed dependency parses from phrase structure parses", | |
| "authors": [ | |
| { | |
| "first": ", M.-C", | |
| "middle": [], | |
| "last": "De Marneffe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Maccartney", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Manning", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "Proceedings of the 5th Language Resources and Evaluation Conference (LREC)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "de Marneffe, M.-C., MacCartney, B., & Manning, C. D. 2006. Generating typed dependency parses from phrase structure parses. In Proceedings of the 5th Language Resources and Evaluation Conference (LREC).", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Universal dependencies", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "De Marneffe", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Manning", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Nivre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Zeman", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2021, | |
| "venue": "Computational Linguistics", | |
| "volume": "47", | |
| "issue": "2", | |
| "pages": "255--308", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "de Marneffe, M. C., Manning, C. D., Nivre, J., & Zeman, D. 2021. Universal dependencies. Computational Linguistics, 47(2), 255-308.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "SUD or Surface-Syntactic Universal Dependencies: An annotation scheme near-isomorphic to UD", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Gerdes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Guillaume", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Kahane", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Perrier", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the Second Universal Dependencies Workshop", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gerdes, K., Guillaume, B., Kahane, S., & Perrier, G. 2018. SUD or Surface-Syntactic Universal Dependencies: An annotation scheme near-isomorphic to UD. In Proceedings of the Second Universal Dependencies Workshop (UDW).", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Improving Surface-syntactic Universal Dependencies (SUD): surface-syntactic relations and deep syntactic features", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Gerdes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Guillaume", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Kahane", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Perrier", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Proceedings of the 18th International Workshop on Treebanks and Linguistic Theories (TLT)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gerdes, K., Guillaume, B., Kahane, S., & Perrier, G. 2019. Improving Surface-syntactic Universal Dependencies (SUD): surface-syntactic relations and deep syntactic features. In Proceedings of the 18th International Workshop on Treebanks and Linguistic Theories (TLT).", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Starting a new treebank? Go SUD!", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Gerdes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Guillaume", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Kahane", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Perrier", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2021, | |
| "venue": "Proceeding of the 6th conference on Dependency Linguistics (Depling)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gerdes, K., Guillaume, B., Kahane, S., & Perrier, G. 2021. Starting a new treebank? Go SUD!. In Proceeding of the 6th conference on Dependency Linguistics (Depling).", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Typometrics: From Implicational to Quantitative Universals in Word Order Typology", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Gerdes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Kahane", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "X", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2021, | |
| "venue": "Glossa: a journal of general linguistics", | |
| "volume": "6", | |
| "issue": "1", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gerdes, K., Kahane, S., & Chen, X. (2021). Typometrics: From Implicational to Quantitative Universals in Word Order Typology. Glossa: a journal of general linguistics, 6(1).", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "A quantitative approach to the morphological typology of language", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Greenberg", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1960, | |
| "venue": "International journal of American linguistics", | |
| "volume": "26", | |
| "issue": "3", | |
| "pages": "178--194", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Greenberg, J. H. 1960. A quantitative approach to the morphological typology of language. International journal of American linguistics, 26(3), 178-194.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Graph Matching and Graph Rewriting: GREW tools for corpus exploration, maintenance and conversion", | |
| "authors": [ | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Guillaume", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2021, | |
| "venue": "Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations (EACL)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "168--175", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Guillaume, B. 2021. Graph Matching and Graph Rewriting: GREW tools for corpus exploration, maintenance and conversion. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations (EACL), 168-175.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "The serial verb construction: Comparative concept and cross-linguistic generalizations", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Haspelmath", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Language and Linguistics", | |
| "volume": "17", | |
| "issue": "3", | |
| "pages": "291--319", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Haspelmath, M. 2016. The serial verb construction: Comparative concept and cross-linguistic generalizations. Language and Linguistics 17(3), 291-319.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "The morph as a minimal linguistic form", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Haspelmath", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2020, | |
| "venue": "Morphology", | |
| "volume": "30", | |
| "issue": "", | |
| "pages": "117--134", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1007/s11525-020-09355-5" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Haspelmath, M. 2020. The morph as a minimal linguistic form. Morphology 30: 117-134. https://doi.org/10.1007/s11525-020-09355-5.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Annotation guidelines of UD and SUD treebanks for spoken corpora", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Kahane", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "B", | |
| "middle": [], | |
| "last": "Caron", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Gerdes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "E", | |
| "middle": [], | |
| "last": "Strickland", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2021, | |
| "venue": "Proceedings of the 20th International Workshop on Treebanks and Linguistic Theories (TLT)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kahane, S., Caron, B., Gerdes, K., & Strickland, E. 2021. Annotation guidelines of UD and SUD treebanks for spoken corpora. In Proceedings of the 20th International Workshop on Treebanks and Linguistic Theories (TLT).", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Coordinate structures in universal dependencies for head-final languages", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Kanayama", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "N", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Han", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Asahara", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Hwang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Miyao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "D" | |
| ], | |
| "last": "Choi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Matsumoto", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the Second Universal Dependencies Workshop", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "75--84", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kanayama, H., Han, N. R., Asahara, M., Hwang, J. D., Miyao, Y., Choi, J. D., & Matsumoto, Y. 2018. Coordinate structures in universal dependencies for head-final languages. In Proceedings of the Second Universal Dependencies Workshop (UDW), 75-84.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "From Fieldwork to Annotated Corpora: The CorpAfroAs Project. Faits de Langue-Les Cahiers n\u00b02", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Mettouchi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Chanard", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "255--265", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Mettouchi, A., & Chanard, C. 2010. From Fieldwork to Annotated Corpora: The CorpAfroAs Project. Faits de Langue-Les Cahiers n\u00b02, 255-265.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Segmentation granularity in dependency representations for Korean", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Park", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proceedings of the Fourth International Conference on Dependency Linguistics (Depling)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "187--196", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Park, J. 2017. Segmentation granularity in dependency representations for Korean. In Proceedings of the Fourth International Conference on Dependency Linguistics (Depling), 187-196.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "The notion of sentence and other discourse units in corpus annotation", | |
| "authors": [ | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Pietrandrea", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Kahane", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Lacheret-Dujour", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Sabio", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Spoken Corpora and Linguistic Studies", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "331--364", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pietrandrea, P., Kahane, S., Lacheret-Dujour, A., & Sabio, F. (2014). The notion of sentence and other discourse units in corpus annotation. In T. Raso, H. Mello, M. Pettorino (eds.), Spoken Corpora and Linguistic Studies, John Benjamins, Amsterdam, 331-364.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "Word-Based or Morpheme-Based? Annotation Strategies for Modern Hebrew Clitics", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "Tsarfaty", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Goldberg", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "Proceedings of the 6th Language Resources and Evaluation Conference (LREC)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tsarfaty, R., & Goldberg, Y. 2008. Word-Based or Morpheme-Based? Annotation Strategies for Modern Hebrew Clitics. In Proceedings of the 6th Language Resources and Evaluation Conference (LREC).", | |
| "links": null | |
| }, | |
| "BIBREF20": { | |
| "ref_id": "b20", | |
| "title": "The Beja language today in Sudan: The state of the art in linguistics", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Vanhove", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2006, | |
| "venue": "Proceedings of the 7th International Sudan Studies Conference", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Vanhove, M. 2006. The Beja language today in Sudan: The state of the art in linguistics. Proceedings of the 7th International Sudan Studies Conference. Bergen: University of Bergen, CD Rom.", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "The Beja Corpus", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Vanhove", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "The CorpAfroAs Corpus of Spoken AfroAsiatic Languages", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1075/scl.68.website" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Vanhove, M. 2014. The Beja Corpus. In Mettouchi, A. and C. Chanard (eds.). The CorpAfroAs Corpus of Spoken AfroAsiatic Languages. http://dx.doi.org/ 10.1075/scl.68.website.", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "title": "Peeters (coll. Les Langues du Monde 9", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Vanhove", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Vanhove, M. 2017. Le Beja. Leuven, Paris: Peeters (coll. Les Langues du Monde 9).", | |
| "links": null | |
| }, | |
| "BIBREF23": { | |
| "ref_id": "b23", | |
| "title": "Hungarian dependency treebank", | |
| "authors": [ | |
| { | |
| "first": "V", | |
| "middle": [], | |
| "last": "Vincze", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Szauter", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Alm\u00e1si", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "M\u00f3ra", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Z", | |
| "middle": [], | |
| "last": "Alexin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Csirik", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Proceedings of the 7th Language Resources and Evaluation Conference (LREC)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Vincze, V., Szauter, D., Alm\u00e1si, A., M\u00f3ra, G., Alexin, Z., & Csirik, J. 2010. Hungarian dependency treebank. In Proceedings of the 7th Language Resources and Evaluation Conference (LREC).", | |
| "links": null | |
| }, | |
| "BIBREF24": { | |
| "ref_id": "b24", | |
| "title": "Character-level Chinese dependency parsing", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "W", | |
| "middle": [], | |
| "last": "Che", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "1326--1336", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Zhang, M., Zhang, Y., Che, W., & Liu, T. 2014. Character-level Chinese dependency parsing. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 1326- 1336.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Derivational affixes (SUD-style)" | |
| }, | |
| "FIGREF1": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Derivational affixes (UD-style)" | |
| }, | |
| "FIGREF2": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "The word-based annotation of the derived word siganfo:j Compare in English:(i)He cleaning the table was impressive. (ii)" | |
| }, | |
| "FIGREF4": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "A Beja independent postposition (UD-style) 'in turn' Figure 11. A Beja enclitic postposition (UD-style) When the postposition complement is a pronoun, it is an enclitic and the postposition precedes its complement (Figure 12): 'in front of her'" | |
| }, | |
| "FIGREF5": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Postposition with an enclitic pronoun (UD-style)" | |
| }, | |
| "FIGREF6": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Topicalized preposed antecedent (UD-style) '(The man was carrying on his shoulder a lamb.) That he had grilled.'" | |
| }, | |
| "FIGREF7": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Preposed antecedent in an afterthought (UD-style)Otherwise the canonical constituent order is used: relative clause -object antecedent -main verb(Figure 18). 'a story like that (happened to me)', lit. a travel that resembles" | |
| }, | |
| "FIGREF8": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Antecedent in canonical word order (UD-style)" | |
| }, | |
| "FIGREF9": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "), and 2 with motion verbs (Figures 21) conjugated at various TAM. '(I went) to wish him a blessed Aid' SVC with verbs of saying (aorist + perfective) (UD-style) 'chase the donkey away!'Figure 21. SVC with motion verbs (both imperfective) (UD-style)" | |
| } | |
| } | |
| } | |
| } |