The Digital Parisian Stage project aims to compile a corpus of plays that will be representative of performances in the theaters of Paris throughout history. The first section has been completed with a random sample of theatrical texts from the period 1800 through 1815, based on the list compiled by Charles Beaumont Wicks (1950), and retrieved from archives. This sampling technique goes beyond the “Principle of Authority ” used for the FRANTEXT corpus, to include playwrights and characters from a wider range of social backgrounds, giving a very different picture of the language. To confirm this broader representation I conducted studies of two groups of morphosyntactic features known to vary with social class : declarative sentence negation and dislocation constructions.
To begin with, a note on the concept of “corpus.” The word means “body” in Latin , and is typically used to refer to a group of texts that constitute a body in the sense of being a coherent whole, although sometimes this coherence can be more imagined and aspirational than real. Corpora have been created and used for hundreds of years, sometimes to study the work of a single author, sometimes of a group of authors, sometimes of a literary canon. The use of the phrases corpus iuris and corpus canonum to refer to a collection of legal texts dates back to the twelfth century (Van Hove 1908). Later corpora were used to group literary texts together for easy reference when writing criticism; Chambers (1728) notes: “We have also a corpus of the Greek poets; and another of the Latin poets.”
Linguists and lexicographers have been compiling digital corpora for analytical papers and dictionaries since the mid-twentieth century. After W. Nelson Francis and Henry Kučera created the Brown Corpus of English (1964), Houghton-Mifflin licensed it as the basis for the American Heritage Dictionary (1969). FRANTEXT was compiled in the 1960s with the goal of creating a new dictionary, the Trésor de la langue française (Imbs 1971). Since that time, other, larger corpora have been created, but FRANTEXT has been the reference corpus for historical studies of the French language.
These corpora are particularly useful for testing diachronic predictions from usage-based linguistic theories, such as the effect of type frequency on linguistic productivity (Bybee 1995). Imagine a French adolescent in the seventeenth century forgetting which negation construction typically goes with the verb cesser and picking the one that seems to go with lots of verbs. This is the type of mechanism assumed by the theory of type frequency.
In order to properly test these theories, however, we need something resembling the language of the past. We have no access to the spontaneous conversations of seventeenth-century adolescents, but instead we can imagine a playwright composing dialogue and verse, reaching into his or her memory for appropriate models. Some of these playwrights may have only looked to earlier playwrights, but others paid attention to the language they heard from their friends, from their servants and on the street.
We have many plays from the past, but we cannot analyze them all at the level we need to test these theories, and they are not all interchangeable. We need a representative sample.
FRANTEXT may have been appropriate for the construction of a dictionary marketed to “the cultivated man” (Imbs 1971: XVIII), but the Principle of Authority introduces a strong bias in favor of elite theater. The Digital Parisian Stage, based on a random sample of all plays that premièred in Paris in the nineteenth century, aims to rectify that bias, offering a broader view of the language of this period that in turn produces more reliable studies of language change .
To compare the Digital Parisian Stage against FRANTEXT, I annotated 22 plays from the Digital Parisian Stage corpus for dislocation and negation features and compared them to the four plays chosen for the FRANTEXT corpus for this period, with striking results. In the Digital Parisian Stage plays, 74% of declarative sentence negations used ne … pas, while in the FRANTEXT plays it was only 50% (p < 0.001).
I chose negation constructions because the study I conducted for my dissertation, The Spread of Change in French Negation (Grieve-Smith 2009), investigated change in negation in the theatrical texts in FRANTEXT. In that study I found that the increase in frequency of ne … pas from the sixteenth through twentieth centuries fit the predictions of Kroch’s (1989) logistic model , and that the logistic model in turn could be explained by the theory of type frequency .
I also looked at left and right dislocation constructions, building on a study where I found a general increase in those constructions in the late twentieth century (Grieve-Smith 2000 ). For left dislocation constructions, 0.760% of non-interrogative sentences used the contrastive topic construction in the Digital Parisian Stage plays, compared to 0.238% in the FRANTEXT plays (p < 0.05, d = 1.04), and 0.113% of non-interrogative sentences (19) used the demonstrative left dislocation (CLD) construction, compared to just one token in FRANTEXT (0.00903%, d = 0.595). The difference in clitic right dislocations was extreme (0.420% for the Digital Parisian Stage plays but only 0.0918% in the FRANTEXT plays, p < 0.01, d = 1.13).
At least one of these differences can be shown to be due to the bias introduced by the Principle of Authority used to compile FRANTEXT. The Digital Parisian Stage corpus contains plays from 9 genres and 11 theaters, while the 4 plays from FRANTEXT are drawn from 3 genres and only 2 theaters, plus 1 closet play. The relative token frequency of ne alone is associated with both the genre of the play and the theater where it was performed, and ne … pas is associated with genre (one-way ANOVA, p < 0.05).
Several of the other negation and dislocation constructions displayed potential associations between corpus and theater, genre or some of the characteristics of the characters (age , gender and social class ). None of these were strong enough to rule out the possibility of sampling error given the size of the sample, but they suggest that once more data is collected for the Digital Parisian Stage corpus, either in the Napoleonic period or later in the nineteenth century, the possibility of sampling error for those factors may be within the typically acceptable range (α = 0.05).
These findings are specific to a short period, 1800–1815, and as such they do not have any immediate bearing on diachronic studies like my dissertation on negation (Grieve-Smith 2009) or my study on dislocation (Grieve-Smith 2000 ). They do suggest that when the 1% sample is complete for the entire nineteenth century, we will see patterns that are similar, but drawn from a more reliable sample that is likely to be closer to informal spontaneous conversation.
It is my hope that these findings will encourage more people to contribute to the Digital Parisian Stage project. For those whose area of focus does not include nineteenth-century Parisian French, I hope this report will encourage them to design and contribute to similarly representative projects in their areas. This will likely include the type of work contributed by Wicks (1950 et seq.), compiling records of language production into a comprehensive catalog that can serve as the sampling frame for a new corpus.
References
- American Heritage Dictionary. 1969. Boston: Houghton Mifflin.
- Bybee, Joan. 1995. Regular Morphology and the Lexicon. Language and Cognitive Processes 10: 425–455.Crossref
- Chambers, Ephraim. 1728. Corpus. In Cyclopædia. London: Chambers.
- Francis, W. Nelson, and Henry Kučera. 1964. A Standard Corpus of Present-Day Edited American English, for Use with Digital Computers. Providence: Brown.
- Grieve-Smith, Angus. 2000. Topicalization and Word Order in Conversational French. Southeastern Conference on Linguistics, Oxford, Mississippi.
- ———. 2009. ...