The Romanian Online Dialect Atlas, phase 2 (RODA II) continues the
development of RODA I, providing a digital version of a hard-copy dialect atlas
(the Noul Atlas lingvistic român. Crişana. or The New Linguistic Romanian
Atlas for the Crişana region by Stan and Uritescu 1997, 2003,
forthcoming mss).
The online atlas includes an online programme for selecting files, searching for data within files, counting the occurrences of particular dialect forms in the data, creating custom maps to display the data and the dialect patterns, and providing tools to analyze the data. There is also a collection of short sound files that provide audio samples of the spoken dialect at various locations.
Romanian is the remaining modern language
in the eastern branch of the Romance language family (the languages derived
from Latin). As such, it shows different
developments than French, Italian, Spanish, etc., which are western Romance
languages. Furthermore, the dialects in Crişana are conservative and show features
that are lost in other Romanian dialects and in standard Romanian. This set of data is very important to our
understanding of how language changes, and is of interest to a wide-range of
scholars, in
With a digital version of the data, it is possible to find patterns in the data that could not be seen if the data had to be examined by hand. In the data set, there are responses to hundreds of linguistic questions (402 in RODA I; over 400 in RODA II) at 120 locations, giving over 96,000 data samples – each of which is a detailed response with many points of interest concerning pronunciation, the shape of words, the choice of words and the arrangement of them into expressions.
Furthermore, RODA offers statistical tools that can compare the differences among locations and portray them as distances on a map. These so-called multidimensional scaling (MDS) maps provide a view of the overall dialect situation that was difficult or impossible to achieve without the use of digitized data and automatic processing algorithms. It has raised some serious theoretical questions about how dialects are distributed in the region, and about geo-linguistic variation in general.
In the second phase of RODA, we intend to
add more data (from volumes III and IV of the original work), refine the tools,
and use the tools to look at the interesting linguistic questions that came from our initial work. We intend to make the data and tools widely
available to scholars, by posting them on the internet at a
Finally, through scholarly publishing and presentations, we reach a wide audience of scholars who can use these tools in their own work, so that we not only share our research findings, but make it possible for others to use our tools and data in ways they find useful.
In phase 2 of the Romanian Online Dialect Atlas (RODA), we will (1) digitize vol. III and IV of the source data (Stan and Uritescu, forthcoming, ms.) consisting of 420 analytic (raw data) maps, and digitize 630 interpretive maps; (2) modify and expand the function of RODA to take into account feedback from users of the initial version of the system; (3) expand the sound files and associated functions to provide both audio samples of the data (up to 15-20 minutes long) from 46 locations across the whole research area, and literary transcriptions of the recordings; (4) use the existing tools to explore the data for linguistic insights that were difficult or impossible to obtain prior to RODA, and (5) in particular, use the multidimensional scaling (MDS) function to examine the homogeneity of the dialect area under study and other linguistic questions, with the greatly expanded amount of data.
The first phase of the Romanian Online Dialect Atlas (RODA) was
supported by SSHRC grant # 410-2003-1076 in 2003-2006, extended to 2008. RODA
is a digital version of an extensive hard-copy atlas of the Crişana region
in North-western
The first two volumes of the hard-copy atlas provide 402 linguistic
maps, showing the detailed responses at 120 locations to questions on body
parts, family and house. Furthermore, most of these maps are matched with
“interpretive” maps that illustrate which regions do or do not have given
features. We digitized the basic data with the help of graduate students in
In an earlier report (Embleton, Uritescu and Wheeler 2004), we addressed the issue of using standard tools (such as commercial databases, Unicode encoding schemes, and existing cartography programmes; see Bird and Simons 2003). We have been forced to use custom software by several factors, including:
· The nature of the data we are dealing with (e.g. the transcription system used in the field greatly extended both standard Romanian and IPA notation);
· The shifting base of what is standard (e.g. the database of choice ten years ago is not the database of choice today);
· The need to make the data accessible and presentable in ways that are familiar to the researchers interested in Romance dialect data.
In the current version of the system, users can select which files they want to search, search them for occurrences of interest in context, count the occurrences, and map either the occurrence or the count by location. In using the system, it became apparent that the user also needed to be able to view the results, and manually adjust them to handle “exceptions”. That function has been added to the system.
Searching and counting are elementary analytic approaches; more sophisticated techniques can also be applied. We provide a statistical function (multidimensional scaling or MDS) that allows the user to view “all” the data as a single map showing the linguistic similarity of the 120 locations. Other techniques also could be added to the system. Our intention has been to provide the data for such techniques, and show the possibilities by implementing one in our system.
Our contributions to research include (1) the support of dialectology (by providing the tools and data that others can use in their case studies); (2) insight into the development of Romance as evidenced by the archaic features of these Daco-Romanian dialects; and (3) the development of quantificational studies of language using statistical techniques such as MDS.
More specifically, we have raised questions (which we hope to answer in phase 2 of RODA) about the dialect divisions in our study area, the existence of dialect boundaries along the Mures and other rivers in the region, and the relation between linguistic continua vs. distinct linguistic areas. As a case study, we can expect to refine the general understanding of how geographical features do and do not create linguistic boundaries.
(1) The digitization of the next two volumes will be done in the
same way as the first two volumes, using an updated version of the data entry
software (suitably changed to get the final data format more directly). Again, we will rely on students and post-doctoral
researchers in
(2) The modification of the system will be a more modest effort than the creation of the system. We are posting the system to the internet, and have publicly invited scholars to use it and provide feedback (in Embleton, Uritescu and Wheeler 2006b, 2007b, 2007c). The feedback will drive the changes we make to the system: we want it to be effective, and to allow them to access the data in ways that are compatible with their normal working methods. We will also rely on our own use of the system to indicate necessary changes. Programming is done in Java, to provide a modern, object-oriented and platform independent application.
(3) The sound files will be extracted from the source data CDs, made into digital audio files. The extracts will be selected to illustrate typical linguistic phenomena from the given location; consideration will also be given to the clarity of the recording, to providing representative samples across the dialect area, and to the feedback on recording length from users of the initial set of sound files. Through the audio clips, RODA I raises the question of the contribution of syntactic features to geolinguistic variation and groupings. Extended audio texts would certainly give researchers the possibility of answering this question.
More generally, our enterprise will lay a solid basis for
investigating a theoretical question in dialectometry
and dialectology, namely the significance of psychoacoustic judgments of
similarity for dialectometric measurements. Some scholars advocate anchoring the notion
of similarity in the perception of dialect users (Heeringa
2004). The procedure, though favoured by some scholars, is considered expensive
(Nerbonne and Kleiweg
2007). In phase 2, RODA will certainly provide the audio data necessary for
such research, data that “one normally does not have” (Nerbonne
and Kleiweg 2007:154).
(4) The research team will use the existing RODA system to explore
relationships in the data that are not easy to see manually. So far, we have been able to challenge or
support several claims about the lenition of vowels, the backing of front
vowels in different contexts, and the palatalization of stops, with substantial evidence about how many lexical examples and
which localities do or do not confirm
previous thinking. We can continue to
take claims from the literature and examine the degree of support for them from
a data set that is much more extensive and tractable than has been available to
researchers before this. Indeed, our extended digitized version of the atlas
could contribute further to the study of many debated linguistic aspects of
Romanian and Romance linguistics, such as the conservation of the plural forms
of nouns ending in schwa or mid-front vowel; the conditional auxiliary and the
past conditional formation; the relation between the two auxiliaries, HABERE and ESSE, and the forms
of the past participle (cf. Uritescu 2007); the area and frequency of archaic
lexical items such the popular descendants of Latin PEDESTER; etc.
(5) In particular, the MDS function allows us to look at several questions that are “impossible” without an automated approach. First, the difference between the dialect groupings revealed by the MDS analysis based on “identical” versus “different” (Embleton, Uritescu and Wheeler 2007b) raises the question of the relation between geo-dialect continua and discrete geolinguistic variation. To answer this question, we need the digitalization of interpretive maps, and an MDS analysis applied to various sets of interpretive maps.
In our current analysis, although the transition area in south-eastern Crişana is well singled out, a second, south-western area is presented as a distinct unity, while the rest of Crişana seems to represent a continuum (Embleton, Uritescu, Wheeler 2007b). This points out two aspects, ignored by previous dialectologists: first, the Crisul Alb river apparently does represent a linguistic border; secondly, contrary to common belief, the large linguistic area north of the Crisul Alb does represent a linguistic unit, in spite of what the analyses of specific features suggest. In fact, our MDS analysis contradicts previous dialectologists’ analyses, in which the unitary character of the linguistic area of Crişana has been doubted (cf. Uritescu 1984b). Is this continuum created only by imbrications of linguistic features analyzed in traditional dialectological studies, or is it also a sort of underlying continuum based on lexical unity? We hope to be able to suggest an answer to this question with the help of MDS analyses of the interpretive maps.
Through the digitalization and MDS analysis of the numerous (610)
interpretive maps, RODA II will also complete the MDS analysis in RODA I and
add a dimension referring to the type of features (phonetic, phonemic,
morphophonemic, morphological, etc.) that most contribute to the constitution
of dialect units. From among other general issues of dialectology and
historical linguistics, the two aspects referred to here, continuum versus
discrete differences in MDS, combined with specific historical research, will
help researchers refine answers to questions concerning relations between socio-geography,
linguistic diffusion and linguistic variation (see Wolfram & Schilling-Estes
2003; Gooskens 2004; Nerbonne
& Kleiweg 2007).
We also wonder if the MDS analysis of linguistic distance in RODA I
does not suggest a grouping of dialects related primarily to a unity of the
basic vocabulary (human body, family and house). A further investigation of
less frequent, less central vocabulary
(agriculture, vegetables, fruit, vineyard, hemp, forest, administration,
etc.) would certainly help us give an answer to this question, since lexical variation
in these terminologies is much greater than in fundamental vocabulary. The role
of infrequent vocabulary in dialect classification is indeed a much debated
aspect in dialectology and dialectometry. While some
scholars (see Manning & Schütze 1999:199, Carver
1987, 1989:17) consider infrequent vocabulary as unreliable for linguistic
structure and for lexical differences in dialectology, others (Goebl 1984, vol. 3, Nerbonne
& Kleiweg 2007) propose a heavier weighting of
infrequent words for the study of dialect similarity and difference.
The inclusion of less frequent vocabulary in
RODA II will certainly contribute, through MDS analysis, to this discussion.
Such an analysis is facilitated also by the lemmatized character of the lexical
interpretive maps in Stan & Uritescu (1996, 2003, forthcoming mss). For the
importance of this aspect, see Nerbonne & Kleiweg 2007 (who develop the analysis proposed by Séguy 1971).
Last, the overlap of dialect groups in
Crişana is much greater than we saw in comparable studies of Finnish and
English dialects (Embleton and Wheeler 1997a, b, 2000) and further analysis of
more data will help clarify this.
Dialectometry is a study of quantitative measures of distance between dialects. It is the “application of the principles of numerical taxonomy to the analysis of dialect data” (Schneider 1984:314), and “aims at the recognition of patterns by means of numerical classification” (Goebl 1984: iii). As Embleton (1993:267) puts it, “it is intended to supplement/complement more traditional methods in dialectology, not to replace them, despite the expectations of some”. It represents in fact “la mise en place de nouveaux instruments méthodiques et … une réorientation méthodologique“ (Goebl 1985:211). Dialectometry, through its quantitative measures, provides an empirical foundation for analyses that otherwise would be just impressionistic interpretations.
There have been several dialectometrical
techniques discussed in the literature, the best known being those of Séguy (and what might be termed a French school of dialectometry centred in
Toulouse; Séguy 1973), of Goebl (especially the use
of ‘choroplethic’ maps, a mapping technique used by
geographers), of various practitioners of hierarchical cluster analysis (such
as Shaw, Linn & Regal, and Klemola), and the dual
scaling technique used by Cichocki. The latter
technique bears resemblance to the method we adopt in the second part of RODA I
and in RODA II, namely multidimensional scaling.
We do not intend to present here our multidimensional scaling method, since it has been presented in other works (Embleton 1987, 1993; Embleton & Wheeler 1997a,b; Chambers 1997, Wheeler 2005) and already applied to two dialectal works by one of the authors of this proposal (see Embleton & Wheeler 1994, 1997a,b, 2000). Multidimensional scaling refers to a collection of statistical techniques for representing the similarities among a set of objects spatially, such as a 2-dimensional map. As a non-technical description puts it, “MDS [multidimensional scaling] provides a means to construct a spatial representation of the [similarities], analogous to constructing a map given distances between pairs of points. Each of the items to be represented is located at a point in an abstract, multidimensional space. MDS brings into focus the global structure of a set of items, rather than local relations among the items themselves” (Davis & Papcun 1987:39).
We construct a matrix of values of the distance measure (D), pair-wise for each pair of dialects being compared; this is in as many dimensions as we have varieties or localities under investigation. Using multidimensional scaling repeatedly, we reduce the number of dimensions to 2, which gives us a ‘linguistic map’ of our dialect space. This map is then compared to the locations of the varieties on the regular ‘geographical map’, which is just a conventional 2-dimensional representation of geographical distance.
As has already been demonstrated (Embleton 1993, Embleton & Wheeler 1997a), the interpretation of the discrepancies between the two types of maps, ‘linguistic’ and ‘geographical’, proves to be extremely interesting for dialectology. To cite yet another dialectologist, “Multidimensional scaling obviously ignores geographic distance and represents only statistical distance. However, it is the comparison between the two types of distance that proves to be one of the most telling of the analysis” (Chambers 1997:291).
In the case of our atlas, such comparisons certainly illuminate in an objective way some of the aspects which impressionistically ‘seem’ to bear importance for the field: the structure of our dialect space, the relations between different linguistic areas, the hierarchy of the dialect areas and of the features that define them (phonetic, morphosyntactic, or lexical), the nature of the transition and isolated areas (for an application of this method to the analysis of a small transitional area in English, see Chambers 1997). Even more, as Embleton (1993) demonstrates, such an analysis could put forward important facts concerning the centres of irradiation of linguistic innovations. We will also be able to discern the role of geographic factors (rivers, mountains, etc.) in the structuring of our dialect space and in the wave-like spread of linguistic innovations.
As to the boundaries, in contrast to the traditional isogloss (“[a line] separating areas which use a particular item from those that [do] not”; Wolfram 1981:49) and isogloss bundle (“the coincidence of a set of isoglosses”; Chambers & Trudgill 1980:109), our approach using multidimensional scaling represents boundaries as the empty space clearly shown on the map, thus also suggesting the ‘fuzzy’ nature of the dialectal boundary.
To cite the opinion of a well-known dialectologist: “Stylizing the geography has not typically been our preference, but perhaps it must become so as we undertake the mapping of more complex linguistic situations” (Chambers 1997:292).
The data and
application system will be posted online (as with RODA I), on a site at
Results will be published in peer-reviewed journals such as the Journal of Quantitative Linguistics, and Dialectologia et Geolinguistica, and presented at the meetings of the associated societies and others, such as: the International Linguistics Association, QUALICO [triennial international conference of the International Quantitative Linguistics Association], International Conference on Methods in Dialectology [triennial], International Congress on Romance Linguistics and Philology [triennial], Digital Resources in the Humanities [annual conference in the UK], ALLC-ACH [annual joint meeting of Association of Literary and Linguistic Computing and Association of Computing in the Humanities]). These venues have proven to draw researchers who are interested in our work.
Furthermore,
the
On going, we expect to include this material in a book on Dialectometry, most likely in the Current Issues in Linguistic Theory series of John Benjamins Inc. (Amsterdam & Philadelphia) [preliminary discussions with the series editor have already been undertaken, and the editor is interested.]
In the
classroom, the material will be presented through Uritescu’s
ongoing undergraduate and graduate teaching of dialectology and Romance dialectology,
and through Wheeler and Embleton’s eventual teaching and general presentations.
Bird, Steven & Gary Simons.
2003. Seven Dimensions of Portability for Language Documentation and
Description. Language. 79. 3. pp. 557-582.
Cassidy, Frederic G. 1978.
On-line mapmaking for the dictionary of American regional English. In Putschke 1978, 107-119.
Chambers, J.K. 1997. Mapping
Transitions. In. Thomas (ed.), pp. 284-293.
Chambers, J.K. and Peter Trudgill. 1980. Dialectology.
Chambers, J.K. and Peter Trudgill. 1998. Dialectology. 2nd ed.
Cichocki, Wladyslaw. 1993. A
Dual Scaling Representation of Phonetic Distances in Acadian French. In Viereck (ed.), pp. 340-355.
Cichocki, Wladyslaw, Louise Péronnet and Rose Mary Babitch.
1997. Atlas Linguistique du
Vocabulaire Maritime Acadien: a final progress report. In Thomas (ed.), pp. 109-119.
Davis, Anthony and George Papcun. 1997. The
structure underlying a semantic domain. In: Manaster-Ramer,
Alexis (ed.), Mathematics of Language.
Embleton, Sheila. 1987.
Multidimensional Scaling as a Dialectometrical
Technique, in Papers from the Eleventh
Annual Meeting of the Atlantic Provinces Linguistic Association, ed. Rose
Mary Babitch. Pp. 33-49.
Embleton, Sheila. 1993.
Multidimensional Scaling as a Dialectometrical
Technique: Outline of a research project, in Contributions to Quantitative Linguistics, Proceedings of the First
Quantitative Linguistics Conference,
Embleton, Sheila & Eric
Wheeler 1994. Dialect Project: Technical
Report.
Embleton, Sheila & Eric
Wheeler. 1997a. Multidimensional Scaling and the SED Data, in The Computer Developed Linguistic Atlas of
Embleton, Sheila & Eric
Wheeler. 1997b. Finnish Dialect Atlas for Quantitative Studies, Journal of Quantitative Linguistics, volume 4, pp. 99-102.
Embleton, Sheila & Eric
Wheeler. 2000. Computerized Dialect Atlas of Finnish: Dealing with Ambiguity, Journal of Quantitative Linguistics,
volume 7, pp. 227-231.
Embleton, Sheila, Dorin Uritescu
& Eric Wheeler. 2002. Online Romanian Dialect Atlas. http://vpacademic.yorku.ca/romanian
Embleton, Sheila, Dorin Uritescu
& Eric Wheeler. 2004. Romanian Online Dialect Atlas. An exploration into
the management of high volumes of complex knowledge in the social sciences and
humanities. Journal of Quantitative
Linguistics. 11.3. 183-192. December 2004.