This article proposes a method that makes possible the linguistic study of textually difficult hand-written materials which are imperfectly preserved. These materials include medieval manuscripts, letters, and legal as well as private documents. With these, the normal treebanking procedure is not sufficient. We present the case of medieval Latin charter texts, i.e., private documents, that 1) are partly fragmentary and 2) exhibit massive use of abbreviations, e.g., chartul for chartulam ‘charter’. In addition, 3) charter texts are highly formulaic and display passages that differ from each other in their language use. It is not possible to ascertain the inflexional endings of most of the fragmentary and abbreviated words, so a method of excluding them from morphological (but not from syntactic) analysis is needed. Moreover, due to the varying degree of formulaicity in certain parts of charter texts, the language of these parts must be studied separately. Therefore, a method of merging two XML layers is introduced. One layer that contains lemmatic, morphological, and syntactic analysis according to the Perseus Latin Dependency Treebank standard is aligned with the other layer that contains textual information (abbreviations, fragmentary words, diplomatic segmentation).

This paper aims to point out a linguistic phenomenon that due to the current stage of research can be analysed only insufficiently with the help of an electronic text corpus. In this way, the paper adds a new aspect to the discussion about historical corpora by tackling the question of how they should be designed in order to be useful for linguistic research on so‐called formulaic patterns. The novelty of the question becomes apparent considering the fact that at present such historical corpora do not exist. In section 1, we define the term formulaic pattern because a clear understanding of this phenomenon is a prerequisite condition for collaborative research of it by historians of language and corpus and computer linguists. Section 2 gives a brief outline of the state of the art in the field of modern formulaic language within the framework of corpus and computer linguistics. Section 3 shows that some well known problems in this area are exacerbated when applied to historical texts. Section 4 presents a possible solution that has been implemented by the HiFoS Researchers' Group at the University of Trier (Germany). Joint research efforts planned with UKP Lab at the TU Darmstadt (section 5) demonstrate that the restrictions posed by historical formulaic patterns are challenges to be overcome, rather than insurmountable obstacles.

In this paper, we explore a novel method to quantify cases of "formulaic syntax" by qualitatively comparing different datasets and adopting different language technology tools with respect to the order of two syntactic constituents (inflected verbs, obliques) in subject relative clauses (Oblique-Verb; Verb-Oblique). We focus our research on the syntactic orders in funerary inscriptions extracted from a small-sized corpus of Latin inscriptions found in the catacombs of Santa Caterina and Santa Mustiola in Chiusi (Italy) compared with a larger repository of Latin inscriptions and treebanks of stages of Latin. Finally, we quantify a "bias" favoring the non-formulaic order (Oblique-Verb) exploring transformer-based deep neural network language models trained with large-scale raw data of Latin. We believe that this type of research can feed the interdisciplinary dialogue between philological studies, theoretical linguistics, and language technology.

Among the many ways in which Latin papyri, a recent and elusive branch of papyrology, can be classified, is the division between papyri where Latin is present as a proper text, with syntactic articulation and a message to convey to the reader; and papyri where Latin is present at only a formulaic level. The former set includes an increasing number of Latin documents on papyrus and ostrakon (currently studied, re-published and published within the project PLATINUM), mostly from Egypt, which testify the needs of Roman citizens and businessmen in that province in the first three centuries of the Empire. The latter set, particularly represented in Late Antiquity till the last centuries of Byzantine Egypt, includes documents where Latin has been often confined to formulae and subscriptions, all produced within provincial bureaux. In this case, the formulae do not do so much as conveying information to the reader at a textual level (this is what the main text does); what they do is determining the status and the relevance of the document in which they are inserted. Three sub-groups can be identified in this set of documents: (1) Greek official letters from provincial authorities to lower ranks, provided with Latin dating formulae in the left and lower margins – a custom which mimics the proceedings of Imperial chanceries producing leges datae; (2) a more miscellaneous subset of letters, complaints and reports in Greek bearing a Latin name, in dative or nominative case, sometimes followed by a rank, as a marginal tag to the document itself; (3) an even more heterogeneous subset, and the object of the present paper: documents sealed by Latin formulae such as legi, bene uale, subscripsi, recognoui, signaui and the like. Whereas in groups (1) and (2) the Latin formula is written by a clerk, in (3) one can often see the very hand of a high-ranking official, with all his graphic peculiarities; his personal – often archaic – tendencies; and his choice of words. Autograph formulae, despite the little amount of text they yield, can uncover several particulars about the education of the subscriber and his relation with his world. One is reminded of a legi in P.Vindob. inv. L 116 (4th-5th AD), still written in ancient Roman cursive; or of the subscribers in the Italian gesta municipalia, whose bene uale still employ the b panse-à-gauche well into the 7th AD. The present paper offers a full survey of such formulae in papyri and ostraka from the two partes Imperii; it attempts at distinguishing those formulae which were informally used and had no strict meaning, from those legally required and unequivocally determined; it analyses the writing practice of the subscribing officers, their graphical education and milieu (Greek or Latin?), and their compliance – or resistance – to the contemporary trends of Roman cursive writing.

This paper aims to point out a linguistic phenomenon that due to the current stage of research can be analysed only insufficiently with the help of an electronic text corpus. In this way, the paper adds a new aspect to the discussion about historical corpora by tackling the question of how they should be designed in order to be useful for linguistic research on so‐called formulaic patterns. The novelty of the question becomes apparent considering the fact that at present such historical corpora do not exist. In section 1, we define the term formulaic pattern because a clear understanding of this phenomenon is a prerequisite condition for collaborative research of it by historians of language and corpus and computer linguists. Section 2 gives a brief outline of the state of the art in the field of modern formulaic language within the framework of corpus and computer linguistics. Section 3 shows that some well known problems in this area are exacerbated when applied to historical text...

Formulaic Language in Encyclopedia of Ancient Greek Language and Linguistics, G. Giannakis ed., Leiden, Brill, 2014, vol. 1, 608-613, Oracular language, ibid. vol. 2, 560-562, Orality and Literacy, ibid. vol. 2, 562-565.

Tools for historical corpus research, and a corpus of Latin We present LatinISE, a Latin corpus for the Sketch Engine. LatinISE consists of Latin works comprising a total of 13 million words, covering the time span from the 2 nd century B. C. to the 21 st century A. D. LatinISE is provided with rich metadata markup , including author, title, genre, era, date and century, as well as book, section, paragraph and line of verses. We have automatically annotated LatinISE with lemma and part-of-speech information. The annotation enables the users to search the corpus with a number of criteria, ranging from lemma, part-of-speech, context, to subcorpora defined chronologically or by genre. We also illustrate word sketches, one-page summaries of a word's corpus-based collocational behaviour. Our future plan is to produce word sketches for Latin words by adding richer morphological and syntactic annotation to the corpus.