{ "paper_id": "W09-0306", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T06:42:56.380337Z" }, "title": "The Development of the Index Thomisticus Treebank Valency Lexicon", "authors": [ { "first": "Barbara", "middle": [], "last": "Mcgillivray", "suffix": "", "affiliation": { "laboratory": "", "institution": "University of Pisa", "location": {} }, "email": "b.mcgillivray@ling.unipi.it" }, { "first": "Marco", "middle": [], "last": "Passarotti", "suffix": "", "affiliation": { "laboratory": "", "institution": "Catholic University of the Sacred Heart Milan", "location": { "country": "Italy" } }, "email": "marco.passarotti@unicatt.it" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "We present a valency lexicon for Latin verbs extracted from the Index Thomisticus Treebank, a syntactically annotated corpus of Medieval Latin texts by Thomas Aquinas. In our corpus-based approach, the lexicon reflects the empirical evidence of the source data. Verbal arguments are induced directly from annotated data. The lexicon contains 432 Latin verbs with 270 valency frames. The lexicon is useful for NLP applications and is able to support annotation.", "pdf_parse": { "paper_id": "W09-0306", "_pdf_hash": "", "abstract": [ { "text": "We present a valency lexicon for Latin verbs extracted from the Index Thomisticus Treebank, a syntactically annotated corpus of Medieval Latin texts by Thomas Aquinas. In our corpus-based approach, the lexicon reflects the empirical evidence of the source data. Verbal arguments are induced directly from annotated data. The lexicon contains 432 Latin verbs with 270 valency frames. The lexicon is useful for NLP applications and is able to support annotation.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "Over the last decades, annotated corpora and computational lexicons have gained an increasing role among language resources in computational linguistics: on the one hand, they are used to train Natural Language Processing (NLP) tools such as parsers and PoS taggers; on the other hand, they are developed through automatic procedures of linguistic annotation and lexical acquisition.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "The relation between annotated corpora and computational lexicons is circular: as a matter of fact, if linguistic annotation of textual data is supported and improved by the use of lexicons, these latter can be induced from annotated data in a corpus-based fashion.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "In the field of cultural heritage and in particular that of classical languages studies, much effort has been devoted throughout the years to the digitization of texts, but only recently have some projects begun to annotate them above the morphological level.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Concerning lexicology and lexicography of classical languages, a long tradition has produced and established many dictionaries, thesauri and lexicons, providing examples from real texts. Nevertheless, nowadays it is possible and indeed necessary to match lexicons with data from (annotated) corpora, and viceversa. This requires the scholars to exploit the vast amount of textual data from classical languages already available in digital format, 1 and particularly those annotated at the highest levels. The evidence provided by the texts themselves can be fully represented in lexicons induced from these data. Subsequently, these lexicons can be used to support the textual annotation itself in a virtuous circle.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "This paper reports on the creation of a valency lexicon induced from the Index Thomisticus Treebank, a syntactically annotated corpus of Medieval Latin texts by Thomas Aquinas. The paper is organised as follows: section 2 describes the available Latin treebanks, their annotation guidelines and gives some specific information on the Index Thomisticus treebank; section 3 deals with the notion of valency, while section 4 describes the state of the art on valency lexicons; section 5 illustrates the procedures of acquisition and representation of our valency lexicon; finally, section 6 draws some conclusions and describes future work.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Latin is a richly inflected language, showing: -discontinuous constituents ('nonprojectivity'): this means that phrasal constituents may not be continuous, but broken up by words of other constituents. An example is the following sentence by Ovid (Metamorphoses, I.1-2): \"In nova fert animus mutatas dicere formas corpora\" (\"My mind leads me to tell of forms changed into new bodies\"). In this sentence, both the nominal phrases \"nova corpora\" and \"mutatas formas\" are discontinuous; -moderately free word-order: for instance, the order of the words in a sentence like \"au-daces fortuna iuvat\" (\"fortune favours the bold\") could be changed into \"fortuna audaces iuvat\", or \"fortuna iuvat audaces\", without affecting the meaning of the sentence. These features of Latin influenced the choice of Dependency Grammars (DG) 2 as the most suitable grammar framework for building Latin annotated corpora like treebanks.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Latin Treebanks", "sec_num": "2" }, { "text": "While since the 1970s the first treebanks were annotated via Phrase Structure Grammar (PSG)based schemata (as in IBM, Lancaster and, later on, Penn treebanks), in the past decade many projects of dependency treebanks development have started, such as the ALPINO treebank for Dutch (Van der Beek et al., 2002) , the Turin University Treebank for Italian (Lesmo et al., 2002) , or the Danish Dependency Treebank (Kromann, 2003) . On the one hand, this is due to the fact that the first treebanks were mainly English language corpora. PSG were a suitable framework for a poorly inflected language like English, showing a fixed word-order and few discontinuous constituents. Later on, the syntactic annotation of moderately free word-order languages required the adoption of the DG framework, which is more appropriate than PSG for such a task. On the other hand, Carroll et al. (1998) showed that inter-annotator agreement was significantly better for dependency treebanks, indicating that phrase structure annotation was requiring too many irrelevant decisions (see also Lin, 1995) .", "cite_spans": [ { "start": 281, "end": 308, "text": "(Van der Beek et al., 2002)", "ref_id": null }, { "start": 353, "end": 373, "text": "(Lesmo et al., 2002)", "ref_id": null }, { "start": 410, "end": 425, "text": "(Kromann, 2003)", "ref_id": null }, { "start": 860, "end": 881, "text": "Carroll et al. (1998)", "ref_id": "BIBREF8" }, { "start": 1069, "end": 1079, "text": "Lin, 1995)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Latin Treebanks", "sec_num": "2" }, { "text": "Although much Latin data is nowadays available in digital format, the first two projects for the development of Latin treebanks have only recently started: namely the Latin Dependency Treebank (LDT) at the Tufts University in Boston (within the Perseus Digital Library) based on texts of the Classical era (Bamman, 2006) , and the Index Thomisticus Treebank (IT-TB) at the Catholic University of the Sacred Heart in Milan, based on the Opera omnia of Thomas Aquinas (Passarotti, 2007) .", "cite_spans": [ { "start": 306, "end": 320, "text": "(Bamman, 2006)", "ref_id": "BIBREF0" }, { "start": 466, "end": 484, "text": "(Passarotti, 2007)", "ref_id": "BIBREF3" } ], "ref_spans": [], "eq_spans": [], "section": "Latin Treebanks", "sec_num": "2" }, { "text": "Taking into account the above mentioned features of Latin, both the treebanks independently chose the DG framework as the most suitable one for data annotation. The same approach was later on followed by a third Latin treebank now available, which is ongoing at the University of Oslo in the context of the PROIEL project (Pragmatic Resources in Old Indo-European Languages): the aim of PROIEL is the syntactic annotation of the oldest extant versions of the New Testament in Indo-European languages, including Greek, Latin, Gothic, Armenian and Church Slavonic (Haug and J\u00f8hndal, 2008) .", "cite_spans": [ { "start": 511, "end": 586, "text": "Greek, Latin, Gothic, Armenian and Church Slavonic (Haug and J\u00f8hndal, 2008)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Latin Treebanks", "sec_num": "2" }, { "text": "Since LDT and IT-TB were the first projects of their kind for Latin, no prior established guidelines were available to rely on for syntactic annotation.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Annotation Guidelines", "sec_num": "2.1" }, { "text": "Therefore, the so-called 'analytical layer' of annotation of the Prague Dependency Treebank (PDT) for Czech (Haji\u010d et al., 1999) was chosen and adapted to specific or idiosyncratic constructions of Latin. These constructions (such as the ablative absolute or the passive periphrastic) could be syntactically annotated in several different ways and are common to Latin of all eras. Rather than have each treebank project decide upon and record each decision for annotating them, LDT and IT-TB decided to pool their resources and create a single annotation manual that would govern both treebanks (Bamman et al., 2007a; Bamman et al., 2007b; .", "cite_spans": [ { "start": 108, "end": 128, "text": "(Haji\u010d et al., 1999)", "ref_id": null }, { "start": 595, "end": 617, "text": "(Bamman et al., 2007a;", "ref_id": "BIBREF2" }, { "start": 618, "end": 639, "text": "Bamman et al., 2007b;", "ref_id": "BIBREF3" } ], "ref_spans": [], "eq_spans": [], "section": "Annotation Guidelines", "sec_num": "2.1" }, { "text": "As we are dealing with Latin dialects separated by 13 centuries, sharing a single annotation manual is very useful for comparison purposes, such as checking annotation consistency or diachronically studying specific syntactic constructions. In addition, the task of data annotation through these common guidelines allows annotators to base their decisions on a variety of examples from a wider range of texts and combine the two datasets in order to train probabilistic dependency parsers.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Annotation Guidelines", "sec_num": "2.1" }, { "text": "Although the PROIEL annotation guidelines are grounded on the same grammar framework as the LDT and IT-TB, they differ in a number of details, some of which are described in Passarotti (forthcoming).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Annotation Guidelines", "sec_num": "2.1" }, { "text": "The Index Thomisticus (IT) by Roberto Busa SJ (1974 -1980 was begun in 1949 and is considered a groundbreaking project in computational linguistics. It is a database containing the Opera omnia of Thomas Aquinas (118 texts) as well as 61 texts by other authors related to Thomas, for a total of around 11 million tokens. The corpus is morphologically tagged and lemmatised.", "cite_spans": [ { "start": 30, "end": 51, "text": "Roberto Busa SJ (1974", "ref_id": null }, { "start": 52, "end": 57, "text": "-1980", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank", "sec_num": "2.2" }, { "text": "Early in the 1970's Busa started to plan a project aimed at both the morphosyntactic disambiguation of the IT lemmatisation and the syntactic annotation of its sentences. Today, these tasks are performed by the IT-TB project, which is part of the wider 'Lessico Tomistico Biculturale', a project whose target is the development of a lexicon from the IT texts. 3 Presently, the size of the IT-TB is 46,456 tokens, for a total of 2,103 parsed sentences excerpted from the Scriptum super Sententiis Magistri Petri Lombardi.", "cite_spans": [ { "start": 360, "end": 361, "text": "3", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank", "sec_num": "2.2" }, { "text": "As outlined above, the notion of valency is generally defined as the number of complements required by a word: these obligatory complements are usually named 'arguments', while the non-obligatory ones are referred to as 'adjuncts'. Although valency can refer to different parts of speech (usually verbs, nouns and adjectives), scholars have mainly focused their attention on verbs, so that the notion of valency often coincides with verbal valency.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency", "sec_num": "3" }, { "text": "Valency is widely used in DG formalisms, but it also figures in PSG-based formalisms like HPSG and LFG.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency", "sec_num": "3" }, { "text": "While Karl B\u00fchler can be considered as the pioneer of the modern theory of valency, 4 Lucien Tesni\u00e8re is widely recognised as its real founder. Tesni\u00e8re views valency as a quantitative quality of verbs, since only verbs constrain both the quantity and the quality (i.e. nouns and adverbs) of their obligatory arguments; through a metaphor borrowed from drama, Tesni\u00e8re classifies dependents into actants (arguments) and circonstants (adjuncts): \"Le noeud verbal [\u2026] exprime tout un petit drame. Comme un drame en effet, il comporte obligatoirement un proc\u00e8s, et le plus souvent des acteurs et des circonstances. Transpos\u00e9s du plan de la r\u00e9alit\u00e9 dramatique sur celui de la syntaxe structurale, le proc\u00e8s, les acteurs et les circonstances deviennent respectivement le verbe, les actants et les circonstants\" (Tesni\u00e8re, 1959: 102). 5", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency", "sec_num": "3" }, { "text": "Arguments can be either obligatory or optional, depending on which sense of the verb is involved. For example, the seem sense of the verb appear requires two obligatory arguments in active clauses, as in the following sentence: \"That lawyer appears to love his work\". Here the second argument (\"to love his work\") cannot be left out without changing the meaning of the verb. On the other hand, optional arguments are recorded into the verbal argument structure itself, althought they may not appear at the clausal level. For instance, in the following sentence the object required by the verb eat is missing, but the sentence is still acceptable: \"He eats (something)\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency", "sec_num": "3" }, { "text": "Optionality can also act at the communicative level as well as at the structural one. For instance, adjuncts can be necessary for communicative intelligibility in particular contexts, as in the following sentence: \"I met James at the Marquee club\", where the locative adverbial (\"at the Marquee club\") is required to answer a question like \"Where did you meet James?\". On the other hand, structural optionality depends on the features of the language and applies at the clausal level. For instance, as a poorly inflected language, English requires the subject of a predicate to be expressed in declarative and interrogative main clauses, so that a sentence like the following is ungrammatical if the subject is missing: \"[I] slept all morning\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency", "sec_num": "3" }, { "text": "Given the so-called \"syntax-semantics interface\" (Levin, 1993), arguments are generally associated with a predicate sense rather than a predicate form, and are structured in sequences called 'subcategorization frames' (SCFs) or 'complementation patterns'. For example, there is a semantic difference between the bill sense and the attack sense of the verb charge in English, as in the following sentences: -(a) \"The hotel charges 80 euros for a night\". -(b) \"The army charged the enemy\".", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency", "sec_num": "3" }, { "text": "In these sentences, the two predicate senses show two different SCFs: -(a) [Subj_NP, Pred, Obj_NP, Obj_PP-for] -(b) [Pred, Obj_NP] Arguments are also selected by verbs according to lexical-semantic properties, called 'selectional preferences' (SPs) or 'selectional restrictions'. For example, a sentence like \"*The train flew to Rome\" is ungrammatical, since it violates and circumstances. Transposed from the dramatic reality to structural syntax, the process, the actors and the circumstances respectively become the verb, the actants and the circumstants\" (our translation). the SP of the verb fly on its subject and can only be accepted in a metaphorical context.", "cite_spans": [ { "start": 116, "end": 130, "text": "[Pred, Obj_NP]", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Valency", "sec_num": "3" }, { "text": "Over the past years, several valency lexicons have been built within different theoretical frameworks: these lexicons have an important role in the NLP community thanks to their wide applications in NLP components, such as parsing, word sense disambiguation, automatic verb classification and selectional preference acquisition.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency Lexicons", "sec_num": "4" }, { "text": "As shown in Ure\u0161ov\u00e1 (2004), a valency lexicon can also help the task of linguistic annotation (as in treebank development), providing annotators with essential information about the number and types of arguments realized at the syntactic level for a specific verb, along with semantic information on the verb's lexical preferences.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency Lexicons", "sec_num": "4" }, { "text": "In the phase of lexicon creation, both intuition-based and corpus-based approaches can be pursued, according to the role played by human intuition and empirical evidence extracted from annotated corpora such as treebanks.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency Lexicons", "sec_num": "4" }, { "text": "For instance, lexicons like PropBank (Kingsbury and Palmer, 2002), FrameNet (Ruppenhofer et al., 2006) and PDT-Vallex (Haji\u010d et al., 2003) have been created in an intuition-based fashion and then checked and improved with examples from corpora.", "cite_spans": [ { "start": 76, "end": 102, "text": "(Ruppenhofer et al., 2006)", "ref_id": null }, { "start": 107, "end": 138, "text": "PDT-Vallex (Haji\u010d et al., 2003)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Valency Lexicons", "sec_num": "4" }, { "text": "On the other side, research in lexical acquisition has recently made available a number of valency lexicons automatically acquired from annotated corpora, such as VALEX (Korhonen, et al., 2006) and LexShem (Messiant et al., 2008) . Unlike the fully intuition-based ones, these lexicons aim at systematically reflecting the evidence provided by data, with very little human intervention. The role of intuition is therefore left to the annotation phase (where the annotator interprets the corpus data), and not extended to the development of the lexicon itself.", "cite_spans": [ { "start": 169, "end": 193, "text": "(Korhonen, et al., 2006)", "ref_id": null }, { "start": 198, "end": 229, "text": "LexShem (Messiant et al., 2008)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Valency Lexicons", "sec_num": "4" }, { "text": "Corpus-based lexicons show several advantages if compared with traditional humandeveloped dictionaries. Firstly, they systematically reflect the evidence of the corpus they were extracted from, while acquiring information specific to the domain of the corpus. Secondly, unlike manually built lexicons, they are not prone to human errors that are difficult to detect, such as omissions and inconsistencies. In addition, such lexicons usually display statistical information in their entries, such as the actual frequency of subcategorization frames as attested in the original corpus. Finally, they are less costly than hand-crafted lexical resources in terms of time, money and human resources.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Valency Lexicons", "sec_num": "4" }, { "text": "While several subcategorization lexicons have been compiled for modern languages, much work in this field still remains to be done on classical languages such as Greek and Latin. Regarding Latin, Happ reports a list of Latin verbs along with their valencies (Happ, 1976: 480-565) . describe a \"dynamic lexicon\" automatically extracted from the Perseus Digital Library, using the LDT as a training set. This lexicon displays qualitative and quantitative information on subcategorization patterns and selectional preferences of each word as it is used in every Latin author of the corpus. Relying on morphological tagging and statistical syntactic parsing of such a large corpus, their approach finds the most common arguments and the most common lexical fillers of these arguments, thus reducing the noise caused by the automatic pre-processing of the data.", "cite_spans": [ { "start": 258, "end": 279, "text": "(Happ, 1976: 480-565)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Valency Lexicons", "sec_num": "4" }, { "text": "We propose a corpus-based valency lexicon for Latin verbs automatically induced from IT-TB data. The automatic procedure allows both the extension of this work to the LDT (thanks to the common annotation guidelines) and the updating of the lexicon as the treebank size increases. First, we automatically extract the arguments of all the occurrences of verbal lemmata in the treebank, along with their morphological features and lexical fillers.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank Valency Lexicon", "sec_num": "5" }, { "text": "In the IT-TB, verbal arguments are annotated using the following tags: Sb (Subject), Obj (Object), OComp (Object Complement) and Pnom (Predicate Nominal); adjuncts are annotated with the tag Adv (Adverbial). The difference between Obj and Adv corresponds to the that between direct or indirect arguments (except subjects) and adjuncts. A special kind of Obj is the determining complement of the object, which is tagged with OComp, such as senatorem in the phrase \"aliquem senatorem facere\" (\"to nominate someone senator\"). Conversely, the determining complement of the subject is tagged as Pnom, as in \"aliquis senator fit\" (\"someone becomes senator\"). 6 In order to retrieve the arguments realised for each verbal occurrence in the treebank, specific database queries have been created to search for the nodes depending on a verbal head through the functional tags listed above.", "cite_spans": [ { "start": 653, "end": 654, "text": "6", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank Valency Lexicon", "sec_num": "5" }, { "text": "The head-dependent relation can be either direct or indirect, since intermediate nodes may intervene. These nodes are prepositions (tag AuxP), conjunctions (tag AuxC) and coordinating or apposing elements (respectively, tags Coord and Apos).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank Valency Lexicon", "sec_num": "5" }, { "text": "For example, see the following sentences: -[1] \"primo determinat formam baptismi;\" 7 (\"at first it determines the form of the baptism;\") -[2] \"ly aliquid autem, et ly unum non determinant aliquam formam vel naturam;\" 8 (\"the 'something' and the 'one' do not determine any form or nature\") Figure 1 reports the tree of sentence [1], where the Obj relation between the verbal head determinat and the dependent formam is direct. Figure 2 shows the tree of sentence [2] . In this tree, two coordinated subjects (aliquid and unum) and two coordinated objects (formam and naturam) depend on the common verbal head determinant through two different Coord nodes (et and vel) 9 . 7 Thomas, Super Sententiis Petri Lombardi, IV, Distinctio 3, Quaestio 1, Prologus, 41-6, 42-2. The edition of the text recorded in the IT is Thomas (1856-1858). 8 Thomas, Super Sententiis Petri Lombardi, III, Distinctio 6, Quaestio 2, Articulus 1, Responsio ad Argumentum 7, 4-5, 6-1. 9 Following PDT-style, the distributed determination aliquam, which modifies both the coordinated objects formam In the case of indirect relation, the intermediate nodes need to be detected and extracted, in order to be inserted into the lexicon as subcategorization structures containing the syntactic roles of the verbal arguments. To represent these structures, we distinguished two major types of them: subcategorization frames (SCFs) and subcategorization classes (SCCs).", "cite_spans": [ { "start": 83, "end": 84, "text": "7", "ref_id": null }, { "start": 462, "end": 465, "text": "[2]", "ref_id": null }, { "start": 671, "end": 672, "text": "7", "ref_id": null } ], "ref_spans": [ { "start": 289, "end": 297, "text": "Figure 1", "ref_id": "FIGREF0" }, { "start": 426, "end": 434, "text": "Figure 2", "ref_id": "FIGREF1" } ], "eq_spans": [], "section": "The Index Thomisticus Treebank Valency Lexicon", "sec_num": "5" }, { "text": "An SCF contains the sequence of functional labels of verbal arguments as they appear in the sentence order, whereas an SCC reports the subcategorization elements disregarding their linear order in the sentence. SCFs and SCCs play a different role in our lexicon. On the one hand, SCFs are very detailed patterns useful for diachronic and/or comparative studies on linear order. On the other hand, SCCs are more general and make the data in the lexicon comparable with the subcategorization structures as usually defined in the literature and in other valency lexicons. For each of these structures we then created the following sub-types, ranging from the most specific to the least specific one. SCF 1 : subcategorization frame marking the full path between the verbal head (referred to as 'V') and each of its argument nodes in the tree. SCF 1 also assigns the same index to those argument nodes linked by coordinating or apposing elements. For instance, the SCF 1 of the verbal and naturam, depends on the coordinating node vel. For more details, see Hajic et al. (1999), 236-238. head determino 10 in sentence [1] is 'V + Obj', while in sentence [2] is '(Coord)Sb_Co (1) + (Coord)Sb_Co (1) + V + (Coord)Obj_Co (2) + (Coord)Obj_Co (2) '. In the latter, the intermediate nodes Coord are in square brackets and indices 1 and 2 link the coordinated nodes. These indices have been adopted in order to disambiguate subcategorization structures where more Obj_Co tags can refer to different verbal arguments. For instance, in a sentence like \"I give X and Y to W and Z\", both the tranferred objects (X and Y) and the receivers (W and Z) are annotated with Obj_Co. Using indices, the subcategorization structure of the verb give in this sentence appears as follows: 'Sb + V + (Coord)Obj_Co (1) + (Coord)Obj_Co (1) + (Coord)Obj_Co (2) + (Coord)Obj_Co (2) '. The indices cannot be applied a priori to subsequent arguments, since Latin, allowing discontinuous constituents, can show cases where coindexed nodes are separated by other lexical items in the linear order.", "cite_spans": [ { "start": 1054, "end": 1083, "text": "Hajic et al. (1999), 236-238.", "ref_id": null }, { "start": 1234, "end": 1237, "text": "(2)", "ref_id": null }, { "start": 1846, "end": 1849, "text": "(2)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank Valency Lexicon", "sec_num": "5" }, { "text": "SCC 1 : the subcategorization class associated with SCF 1 . The SCC 1 of the verb determino in", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank Valency Lexicon", "sec_num": "5" }, { "text": "[1] is '{Obj}', while in [2] is '{(Coord)Sb_Co (1) , (Coord)Sb_Co (1) , (Coord)Obj_Co (2) , (Co- ord)Obj_Co (2) }'.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank Valency Lexicon", "sec_num": "5" }, { "text": "SCF 2 : a subcategorization frame containing only the labels and the indices of the arguments, but not the full path. So, the SCF 2 of determino in [1] is 'V + Obj', while in [2] is 'Sb_Co (1) + Sb_Co (1) + V + Obj_Co (2) + Obj_Co (2) '. SCC 2 : the subcategorization class associated with SCF 2 . For determino, this is '{Obj}' in [1] and '{Sb_Co (1) , Sb_Co (1) , Obj_Co (2) , Obj_Co (2) }' in [2] . SCC 3 : a subcategorization frame containing only the argument labels. The SCC 3 of determino is '{Obj}' in [1] and '{Sb, Obj}' in [2] , showing that in this sentence determino is used as a biargumental verb, regardless of the number of lexical fillers realised for each of its arguments at the surface level.", "cite_spans": [ { "start": 332, "end": 335, "text": "[1]", "ref_id": null }, { "start": 348, "end": 351, "text": "(1)", "ref_id": null }, { "start": 396, "end": 399, "text": "[2]", "ref_id": null }, { "start": 510, "end": 523, "text": "[1] and '{Sb,", "ref_id": null }, { "start": 524, "end": 536, "text": "Obj}' in [2]", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "The Index Thomisticus Treebank Valency Lexicon", "sec_num": "5" }, { "text": "Presently, the size of the IT-TB valency lexicon is 432 entries (i.e. verbal lemmata, corresponding to 5966 wordforms), with 270 different SCF 1 s. In the near future, the lexicon will be enriched with valency information for nouns and adjectives.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "The corpus-based approach we followed induces verbal arguments directly from annotated data, where the arguments may be present or not, 10 Determino is the lemma of both the wordforms determinat (sentence [1]) and determinant (sentence [2]).", "cite_spans": [ { "start": 136, "end": 138, "text": "10", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "depending on the features of the texts. Therefore, the lexicon reflects the empirical evidence given by the data it was extracted from, encouraging linguistic studies on the particular language domain of our corpus.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "In addition to the syntactic information reported in the different types of SCFs and SCCs, it is possible at each stage to include both the morphological features and the lexical fillers of verbal arguments, helping define verbal selectional preferences.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "The lexicon may also be useful for improving the performance of statistical parsers, enriching the information acquired by parsers on verbal entries. On the other hand, moving from parser performance to lexicon development, the lexicon can be induced from automatically parsed texts when an accurate parsing system is available.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "The syntactic and lexical data recorded in the lexicon are also important in further semantic NLP applications, such as word sense disambiguation, anaphora and ellipsis resolution, and selectional preference acquisition. Following a widespread approach in valency lexicons, a close connection between valency frames and word senses will be followed in the description of lexicon entries: this means that each headword entry of our lexicon will consist of one or more SCFs and SCCs, one for each sense of the word.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "We plan to make the lexicon available online through a graphical interface usable also during the annotation procedures, as has been already done for the PDT via the tree editor TrEd. 11 In this way, the consistency of the annotation process can be tested and enforced thanks to the information stored in the lexicon.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "In order to test the accuracy of our system, it will be also necessary to evaluate the quality of our valency lexicon against the Perseus \"dynamic lexicon\", Happ's list and other existing resources for Latin, such as traditional dictionaries and thesauri. A comparison with the lexicon by Perseus is also very interesting in a contrastive diachronic perspective, as it may show important linguistic differences between Classical and Medieval Latin.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "TrEd is freely available at http://ufal.mff.cuni.cz/~pajas/tred/.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and future work", "sec_num": "6" }, { "text": "See, for instance, the Perseus Digital Library(Crane et al., 2001), or data repositories such as LASLA(Denooz, 1996).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "With Tesni\u00e8re (1959) as a common background, there are many different current DG flavours. See for instance the following: Dependency Unification Grammar(Hellwig, 1986), Functional Generative Description(Sgall, Haji\u010dov\u00e1 and Panevov\u00e1, 1986), Meaning Text Theory(Mel'\u010duk, 1988), Word Grammar(Hudson, 1990).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "http://itreebank.marginalia.it.4 In the Sprachtheorie, he writes that \"die W\u00f6rter einer bestimmten Wortklasse eine oder mehrere Leerstellen um sich er\u00f6ffnen, die durch W\u00f6rter bestimmter anderer Wortklassen ausgef\u00fcllt werden m\u00fcssen\"(B\u00fchler, 1934: 173) (\"words of a certain word-class open up around themselves one or several empty spaces that have to be filled by words of certain other word-classes\"; our translation). 5 \"The verbal node expresses a whole little drama. As a drama, it implies a process and, most of the times, actors", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "As in the PDT, all of the syntactic tags can be appended with a suffix in the event that the given node is member of a coordinated construction (_Co), an apposition (_Ap) or a parenthetical statement (_Pa).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [ { "text": "We would like to thank Paolo Ruffolo for his help in designing the database architecture. (LASLA). \u00ab Le M\u00e9di\u00e9viste et l 'ordinateur \u00bb, 33, 14-20. Jan Haji\u010d, Jarmila Panevov\u00e1, Eva Bur\u00e1\u0148ov\u00e1, Zde\u0148ka Ure\u0161ov\u00e1 and Alla B\u00e9mov\u00e1. 1999 ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Acknowledgments", "sec_num": null }, { "text": "Processing (ICON 2002 ), Vikas Publ. House, New Delhi, 61-70. Beth Levin. 1993 ", "cite_spans": [ { "start": 11, "end": 21, "text": "(ICON 2002", "ref_id": null }, { "start": 22, "end": 78, "text": "), Vikas Publ. House, New Delhi, 61-70. Beth Levin. 1993", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "annex", "sec_num": null } ], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "The Design and Use of Latin Dependency Treebank", "authors": [ { "first": "David", "middle": [], "last": "Bamman", "suffix": "" } ], "year": 2006, "venue": "TLT 2006. Proceedings of the Fifth Workshop on Treebanks and Linguistic Theories", "volume": "", "issue": "", "pages": "67--78", "other_ids": {}, "num": null, "urls": [], "raw_text": "David Bamman. 2006. The Design and Use of Latin Dependency Treebank. In Jan Haji\u010d and Joakim Nivre (eds.), TLT 2006. Proceedings of the Fifth Workshop on Treebanks and Linguistic Theories. December 1-2, 2006, Prague, Czech Republic, In- stitute of Formal and Applied Linguistics, Prague, Czech Republic, 67-78.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Building a Dynamic Lexicon from a Digital Library", "authors": [ { "first": "David", "middle": [], "last": "Bamman", "suffix": "" }, { "first": "Gregory", "middle": [], "last": "Crane", "suffix": "" } ], "year": 2008, "venue": "Proceedings of the 8th ACM/IEEE-CS Joint Conference on Digital Libraries", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "David Bamman and Gregory Crane. 2008. Building a Dynamic Lexicon from a Digital Library. In Pro- ceedings of the 8th ACM/IEEE-CS Joint Confer- ence on Digital Libraries (JCDL 2008), Pittsburgh.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Guidelines for the Syntactic Annotation of Latin Treebanks, \u00abTufts University Digital Library", "authors": [ { "first": "David", "middle": [], "last": "Bamman", "suffix": "" }, { "first": "Marco", "middle": [], "last": "Passarotti", "suffix": "" }, { "first": "Gregory", "middle": [], "last": "Crane", "suffix": "" }, { "first": "Savina", "middle": [], "last": "Raynaud", "suffix": "" } ], "year": 2007, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "David Bamman, Marco Passarotti, Gregory Crane and Savina Raynaud. 2007a. Guidelines for the Syntac- tic Annotation of Latin Treebanks, \u00abTufts Univer- sity Digital Library\u00bb. Available at: http://dl.tufts.edu/view_pdf.jsp?urn=tufts:facpubs:d bamma01-2007.00002.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "A Collaborative Model of Treebank Development", "authors": [ { "first": "David", "middle": [], "last": "Bamman", "suffix": "" }, { "first": "Marco", "middle": [], "last": "Passarotti", "suffix": "" }, { "first": "Gregory", "middle": [], "last": "Crane", "suffix": "" }, { "first": "Savina", "middle": [], "last": "Raynaud", "suffix": "" } ], "year": 2007, "venue": "Proceedings of the Sixth International Workshop on Treebanks and Linguistic Theories", "volume": "1", "issue": "", "pages": "1--6", "other_ids": {}, "num": null, "urls": [], "raw_text": "David Bamman, Marco Passarotti, Gregory Crane and Savina Raynaud. 2007b. A Collaborative Model of Treebank Development. In Koenraad De Smedt, Jan Haji\u010d and Sandra K\u00fcbler (eds.), Proceedings of the Sixth International Workshop on Treebanks and Linguistic Theories. December 7-8, 2007, Ber- gen, Norway, Northern European Association for Language Technology (NEALT) Proceedings Se- ries, Vol. 1, 1-6.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "The annotation guidelines of the Latin Dependency Treebank and Index Thomisticus Treebank. The treatment of some specific syntactic constructions in Latin", "authors": [ { "first": "David", "middle": [], "last": "Bamman", "suffix": "" }, { "first": "Marco", "middle": [], "last": "Passarotti", "suffix": "" }, { "first": "Roberto", "middle": [], "last": "Busa", "suffix": "" }, { "first": "Gregory", "middle": [], "last": "Crane", "suffix": "" } ], "year": 2008, "venue": "Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC 2008)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "David Bamman, Marco Passarotti, Roberto Busa and Gregory Crane. 2008. The annotation guidelines of the Latin Dependency Treebank and Index Thomis- ticus Treebank. The treatment of some specific syntactic constructions in Latin. In Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC 2008). May 28- 30, 2008, Marrakech, Morocco, European Lan- guage Resources Association (ELRA), 2008.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Sprachtheorie: die Darstellungsfunktion der Sprache", "authors": [ { "first": "Karl", "middle": [], "last": "B\u00fchler", "suffix": "" } ], "year": 1934, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Karl B\u00fchler. 1934. Sprachtheorie: die Darstellungs- funktion der Sprache, Jena: Gustav Fischer, Stutt- gart.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Index Thomisticus: sancti Thomae Aquinatis operum omnium indices et concordantiae, in quibus verborum omnium et singulorum formae et lemmata cum suis frequentiis et contextibus variis modis referuntur quaeque / consociata plurium opera atque electronico IBM automato usus digessit Robertus Busa SJ, Frommann-Holzboog", "authors": [ { "first": "Roberto", "middle": [], "last": "Busa", "suffix": "" } ], "year": 1974, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Roberto Busa. 1974-1980. Index Thomisticus: sancti Thomae Aquinatis operum omnium indices et con- cordantiae, in quibus verborum omnium et singu- lorum formae et lemmata cum suis frequentiis et contextibus variis modis referuntur quaeque / con- sociata plurium opera atque electronico IBM automato usus digessit Robertus Busa SJ, From- mann-Holzboog, Stuttgart-Bad Cannstatt.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Drudgery and deep thought: Designing a digital library for the humanities", "authors": [ { "first": "Gregory", "middle": [ "R" ], "last": "Crane", "suffix": "" }, { "first": "Robert", "middle": [ "F" ], "last": "Chavez", "suffix": "" }, { "first": "Anne", "middle": [], "last": "Mahoney", "suffix": "" }, { "first": "Thomas", "middle": [ "L" ], "last": "Milbank", "suffix": "" }, { "first": "Jeff", "middle": [ "A" ], "last": "Rydberg-Cox", "suffix": "" }, { "first": "David", "middle": [ "A" ], "last": "Smith", "suffix": "" }, { "first": "Clifford", "middle": [ "E" ], "last": "Wulfman", "suffix": "" } ], "year": 2001, "venue": "Communications of the ACM", "volume": "44", "issue": "5", "pages": "34--40", "other_ids": {}, "num": null, "urls": [], "raw_text": "Gregory R. Crane, Robert F. Chavez, Anne Mahoney, Thomas L. Milbank, Jeff A. Rydberg-Cox, David A. Smith and Clifford E. Wulfman. 2001. Drudg- ery and deep thought: Designing a digital library for the humanities. In Communications of the ACM, 44(5), 34-40.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Parser Evaluation: a Survey and a New Proposal", "authors": [ { "first": "John", "middle": [], "last": "Carroll", "suffix": "" }, { "first": "Ted", "middle": [], "last": "Briscoe", "suffix": "" }, { "first": "Antonio", "middle": [], "last": "Sanfilippo", "suffix": "" } ], "year": 1998, "venue": "Proceedings of the First International Conference on Language Resources and Evaluation (LREC 1998)", "volume": "", "issue": "", "pages": "447--454", "other_ids": {}, "num": null, "urls": [], "raw_text": "John Carroll, Ted Briscoe and Antonio Sanfilippo. 1998. Parser Evaluation: a Survey and a New Pro- posal. In Proceedings of the First International Conference on Language Resources and Evalua- tion (LREC 1998). May 28-30, 1998, Granada, Spain, 447-454.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "La banque de donn\u00e9es du laboratoire d'analyse statistique des langues anciennes", "authors": [ { "first": "Joseph", "middle": [], "last": "Denooz", "suffix": "" } ], "year": 1996, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Joseph Denooz. 1996. La banque de donn\u00e9es du labo- ratoire d'analyse statistique des langues anciennes", "links": null } }, "ref_entries": { "FIGREF0": { "type_str": "figure", "uris": null, "num": null, "text": "Tree of sentence [1]" }, "FIGREF1": { "type_str": "figure", "uris": null, "num": null, "text": "Tree of sentence [2]" } } } }