Languages and Literatures

Many varieties of French have changed over the years from expressing predicate negation (Geurts 1998) with ne alone, to the embracing construction ne … pas, and then to postverbal pas alone (Jespersen 1917). When the increase in the frequency of ne … pas over time is plotted on a graph, it takes the S shape of the logistic function (Kroch 1989). Bybee and Thompson (1997) note that "the type frequency of a pattern determines its degree of productivity," but "high frequency forms with alternations resist analogical leveling.” These two observations provide an explanation for the logistic progression observed by Kroch (1989). Following Lotka (1925) and Volterra (1926), we can extend this model to take into account the competition between constructions to express the same function. To test these models, I have compiled a corpus of French theatrical texts from the twelfth to the twentieth century. The logistic function accurately models the use of ne … pas in these texts (R2 = 0.899), but the Lotka-Volterra model predicts the post-1600 changes in preverbal ne alone and embracing ne … pas and ne … point with even greater accuracy (r = 0.948 and 0.978).

Negation has long been a favorite topic of study for linguists, and French negation especially so, because of its distinctiveness among the more widely studied languages.  In this study I will apply the principles of grammaticization theory, as regards in particular the effects of frequency (Bybee and Thompson 1997), to explain some of the events in the history of French negation.  I will analyze a series of texts, mostly plays, drawn from the body of French literature, to show that the shift from preverbal to postverbal negation is an example of grammaticization, and that the spread of the “embracing negation” ne...pas is an example of analogical extension driven by high type frequency, and resisted by high token frequency constructions.

The changes that I will be discussing are a small part of the larger history of the French language, and in later chapters we will see how strongly they were affected by this history. We have over a thousand years of records to draw on, from which I will be examining a selection of theatrical texts. In this chapter I will give a brief overview of the sociopolitical history of the language. I will then discuss the overall morphosyntactic history of negation in French, beginning with its antecedents in Latin and continuing on through the present day; in a later chapter I will discuss the current understanding of the semantic and pragmatic evolution of the term. Where possible I will illustrate the discussion with examples from the corpus I have collected, and if necessary I will supplement these examples with excerpts from more recent texts.

Over the years, a number of models have been proposed to account for widespread observations of S-shaped curves in language change. In all instances there are at least two competing variants for expressing the same meaning or function, and the frequency of one rises slowly at first, then more quickly, and finally tapers off in the direction of full use. Niyogi and Berwick (1997) and others propose a model where a child chooses between two competing grammars during acquisition. In earlier work (Grieve-Smith 2009) I suggest two models, both based on type frequency. Most recently, Baxter et al. (2009) put forth a detailed model with differential weighting, and Blythe and Croft (2012) apply it to S-curves.

Croft (2000) and Bybee (2010) compile data from several studies showing that the learning patterns of small children do not correspond to patterns of language change. Lifelong models like those of Grieve-Smith (2009) and Blythe and Croft (2012) are more consistent with this data. These three lifelong models are in fact compatible with each other, and we can make the choice of model based on the situation at hand. In my study (Grieve-Smith 2009), I begin with the logistic model proposed by Kroch (1989), and motivate it through the relative type frequencies of the variants. I then observe that in its original field of population modeling, the logistic was extended by Lotka (1925) and Volterra (1926), who both added a term αij to indicate an advantage that term i may have over term j. Conversely, Blythe and Croft (2012) focus on the differential weighting that language users may apply to the two variants, with a "sampling rule" f(u) that allows the model to take type frequency into account.

All three models produce satisfactory S-curves. Grieve-Smith (2009) examined corpus data from the evolution of ne … pas in French theatrical texts and found that the logistic model (R2 = 0.899) and the Lotka-Volterra model (r = 0.978). fit the data closely. Blythe and Croft (2012) do not test their model's predictions against the data, but they do use their model to generate idealized data and report that it generates "an S-shaped trajectory." They observe that it is not necessary to incorporate frequency effects because "the simplest choice of the sampling rule f(u) that implements replicator selection is sufficient to obtain an S-curve," but their model does not preclude taking frequency into account.

When examining S-curves in language change, then, we have three models to choose from. The logistic (Kroch 1989), which only uses frequency data, is the simplest and the easiest to implement, and will be sufficient for many datasets. The Lotka-Volterra model (Grieve-Smith 2009), which assigns a different competitive advantage to each variant, can be used when focusing on such advantages. The interactive model of Blythe and Croft (2012) can be used when we want to examine the interactions between language users in greater detail.

L’évolution de la négation en français est un des changements les plus étudiés dans la syntaxe (Schweighäuser 1852, Jespersen 1917, Yvon 1962, Martineau et Mougeon 2003 entre autres). C’est un des exemples les plus connus de grammaticalisation (Detges et Waltereit 2002, Hopper et Traugott 2003, Rostila 2006). Bybee et Thompson (2000) ont observé que la productivité d’une construction dépend largement de sa fréquence de type, et la résistance d’un item lexical face à un changement analogique dépend de sa fréquence d’occurrence. Mais à part Rostila (2006), ces théories des effets de fréquence n’ont pas encore été appliquées à l’évolution de la négation.

A corpus of texts needs to be based on a representative sample to justify generalizing any findings beyond the individual text, but there are numerous challenges to this (Biber 1993).  For historical corpora we can add the challenges of working with archival texts.  One way to deal with some of these challenges is to choose a well-defined sampling frame that lends itself to corpus compilation, while allowing the reader to judge its fitness for the task.

One promising sampling frame is Wicks’s exhaustive list of all plays that premiered in public in Paris in the nineteenth century, totaling over 30,000.  The first volume (Wicks 1950), covering the years 1800-1815, contains over 3100 plays.  I have extracted a random sample of thirty-one plays, (one percent) from this first volume, and obtained copies of twenty-four of those.

We may well expect a corpus based on a random sample to give us a different picture from previous corpora based on judgments of canonical worth, such as FRANTEXT (2014).  This can be seen by brief investigations of well-known variables. For example, in the four theatrical texts for this period in FRANTEXT, 49% of negated sentences on average used ne … pas, 21% ne … point, and 30% ne alone. In four plays chosen at random from the Napoleonic corpus, there is 87% ne … pas (p = 0.0002), 4% ne … point (p = 0.0015), and 9% ne alone (p = 0.00007).  The fact that these are random samples even justifies the use of p values in this instance.

We need more representative corpora in order to have confidence when generalizing our results, but there are numerous challenges (Biber 1993), and for historical corpora we can add the challenges of working with archival texts.  One way to deal with some of these challenges is to choose a well-defined sampling frame that lends itself to corpus compilation, while allowing the reader to judge its fitness for the task.
One promising sampling frame is Wicks’s exhaustive list of all plays that premiered in public in Paris in the nineteenth century, totaling over 30,000.  The first volume (Wicks 1950), covering the years 1800-1815, contains over 3100 plays.  I have extracted a random sample of thirty-one plays, (one percent) from the first volume, and obtained copies of twenty-four of those.
We may well expect a corpus based on a random sample to give us a different picture from previous corpora based on judgments of canonical worth, such as FRANTEXT (2014).  This can be seen by brief investigations of well-known variables. For example, on average in the four theatrical texts for this period in FRANTEXT, 49% of negated declarative sentences used ne … pas, 21% ne … point, and 30% ne alone. In seven texts chosen at random from the Napoleonic corpus, we find on average 73% ne … pas (p < 0.01), 14% ne alone (p < 0.01), and 14% ne … point.  This difference is likely due to the fact that three of the four FRANTEXT plays feature aristocratic characters, while many of the randomly selected plays feature characters who are servants, peasants and artisans.

The Termolator is an open-source high-performing terminology extraction system, available on Github. The Termolator combines several different approaches to get superior coverage and precision. The in-line term component identifies potential instances of terminology using a chunking procedure, similar to noun group chunking, but favoring chunks that contain out-of-vocabulary words, nominalizations, technical adjectives, and other specialized word classes. The distributional component ranks such term chunks according to several metrics including: (a) a set of metrics that favors term chunks that are relatively more frequent in a "foreground" corpus about a single topic than they are in a "background" or multi-topic corpus; (b) a well-formedness score based on linguistic features; and (c) a relevance score which measures how often terms appear in articles and patents in a Yahoo web search. We analyse the contributions made by each of these components and show that all modules contribute to the system's performance, both in terms of the number and quality of terms identified. This paper expands upon previous publications about this research and includes descriptions of some of the improvements made since its initial release. This study also includes a comparison with another terminology extraction system available on-line, Termostat (Drouin, 2003). We found that the systems get comparable results when applied to small amounts of data: about 50% precision for a single foreground file (Einstein's Theory of Relativity). However, when running the system with 500 patent files as foreground, Termolator performed significantly better than Termostat. For 500 refrigeration patents, Termolator got 70% precision vs. Termostat's 52%. For 500 semiconductor patents, Termolator got 79% precision vs. Termostat's 51%.

The Termolator is an open-source high-performing terminology extraction system, available on Github. The Termolator combines several different approaches to get superior coverage and precision. The in-line term component identifies potential instances of terminology using a chunking procedure, similar to noun group chunking, but favoring chunks that contain out-of-vocabulary words, nominalizations, technical adjectives, and other specialized word classes. The distributional component ranks such term chunks according to several metrics including: (a) a set of metrics that favors term chunks that are relatively more frequent in a "foreground" corpus about a single topic than they are in a "background" or multi-topic corpus; (b) a well-formedness score based on linguistic features; and (c) a relevance score which measures how often terms appear in articles and patents in a Yahoo web search. We analyse the contributions made by each of these components and show that all modules contribute to the system's performance, both in terms of the number and quality of terms identified. This paper expands upon previous publications about this research and includes descriptions of some of the improvements made since its initial release. This study also includes a comparison with another terminology extraction system available on-line, Termostat (Drouin, 2003). We found that the systems get comparable results when applied to small amounts of data: about 50% precision for a single foreground file (Einstein's Theory of Relativity). However, when running the system with 500 patent files as foreground, Termolator performed significantly better than Termostat. For 500 refrigeration patents, Termolator got 70% precision vs. Termostat's 52%. For 500 semiconductor patents, Termolator got 79% precision vs. Termostat's 51%.

This paper deals with there + be constructions in English and the yes construction in Hebrew. The purpose of this study is to identify the stages and strategies being used within child second language acquisition (L2) by examining the interlanguage (IL) of English speakers studying Hebrew as their second language. In particular, this study investigates the use of the Hebrew yes construction in written compositions of native English speakers studying Hebrew in grades 4-7 (ages 9-13) at a private Jewish day school in Chicago 11.