2.1. The Diminutive Suffix -la

Let us first illustrate the morphological diminutive, which is realized by the suffix -la in East Franconian. It is well known that diminutive marking in Standard German turns mass nouns into count nouns. The same holds for Upper German versions of Standard German -chen in 1b, such as East Franconian -la in 2b.

1. (1)

   

1. (2)

   

As soon as the diminutive suffixes -chen and -la are present, as in 1b and 2b, respectively, the noun must be interpreted as a count noun (Jurafsky Reference Jurafsky1996). In what follows, we capitalize on this central fact and abstract away from further semantic and/or pragmatic functions that the diminutive might have. Intuitively, at least in Standard German, the diminutive shrinks the size of its noun referent, but here we start with the assumption that this additional meaning component might vary from language to language, and from variety to variety. We hypothesize that in Upper German dialects such as East Franconian, the high frequency of the use of diminutive forms might actually correspond to semantic bleaching (for further semantic/pragmatic effects of diminutives from a crosslinguistic perspective, see Fortin Reference Fortin2011, Parzuchowski et al. Reference Parzuchowski, Bocian and Gygax2016, and our discussion in section 5 below).^{Footnote 1}

Accordingly, as soon as one focuses only on the productive pattern of diminutive marking on mass nouns, it becomes clear that diminutives can be analyzed along the lines of other numeral classifiers such as Glas ‘glass’ in so-called “counting constructions” (Wiese & Maling 2005):

1. (3)

   

Since numeral classifiers such as Glas share this basic semantic feature of turning mass into count nouns, many syntactic analyses treat them as similar and propose structural analyses such as in 4 in order to account for this parallel (see Wiltschko Reference Wiltschko2006, Ott Reference Ott2011, and many others). The category n is the “light noun”—a category that has been used for a variety of phenomena at the syntax-morphology interface (Marantz Reference Marantz1997), and especially for classifier systems of East Asian languages (for example, Kishimoto Reference Kishimoto2000, Hiraiwa Reference Hiraiwa2016, Reference Hiraiwa2018).^{Footnote 2}

1. (4)

   

In 4, diminutive suffixes, along with further numeral classifiers, can thus be analyzed as light nouns. This approach is in line with an insight that the only semantic effect of the morphological diminutive that seems to hold across languages and varieties is that it specifies a countable entity of the concept expressed by a noun (as in das Bier _mass → das Bier-chen _count; Bale & Barner 2018), which then also could be pluralized (as in die Bier-chen _count). This process can be modeled either using the exoskeletal approach outlined in 4, analyzed within the level of semantics through a process of lambda abstraction, as proposed by Jurafsky (Reference Jurafsky1996), or through a process of morphological composition that signals a conceptual effect (Wiese Reference Wiese2006).

Under either account, the smallest common denominator of nominal diminutives is that they refer to something like “one piece of N” at the conceptual level, and it remains to be seen whether or not diminutive suffixes also have interpretive effects beyond this basic component of conceptualization—for instance, by additionally modifying the size of the noun referent (that is, one small piece of N; for example, Schneider Reference Schneider2013). Before investigating this question in more detail in sections 3 and 4 below, let us first turn to another diminutive strategy in East Franconian whose distribution with mass versus count nouns follows a similar pattern and can thus be profitably compared with diminutive suffixes: the quantifying construction a weng a N.

2.2. The Quantifying Construction a Weng a.

The quantifier weng is the East Franconian version of Standard German wenig ‘few, little’. Interestingly, this quantifier can occur in a construction with indefinite determiner doubling, as in 5, which is a signature property of Upper German dialects, especially of Bavarian (Brandner Reference Brandner2008, Kallulli & Rothmayr 2008, Leu Reference Leu2008, Reference Leu2015).

1. (5)

   

As already pointed out in the literature, the doubling of definite determiners in Upper German varieties is restricted. That is, it can only occur with a certain set of lexical elements such as ganz ‘very, completely’. An example would be the following Austro-Bavarian case from Kallulli & Rothmayr 2008:101 in 6a. However, it has also been noted, in the very same literature, that there are differences in acceptability between Upper German varieties in Austria and Germany, and, according to our judgment, the corresponding example in 6b is not possible in East Franconian.

1. (6)

   

‘You are the very clever/the cleverest ones.’

Although judgments may thus vary regarding doubling of the definite determiner as such, it is undisputed that co-occurrence of the definite determiner with weng is only possible when the latter functions as a quantifier under the mass-noun reading of the noun referent, as in 7a; the doubling of definite determiners with weng is ungrammatical, as in 7b,c

1. (7)

   

We thus conclude that doubling of a determiner in East Franconian in a construction that also contains weng is only possible with the indefinite determiner (see 5 above). Focusing on indefinite determiner configu-rations, we now observe that without doubling of the indefinite determiner, weng can again only occur with mass nouns:

1. (8)

   

Given our discussion in section 2.1, this predicts that indefinite determiner doubling + weng should also be acceptable with the diminutive suffix -la because this suffix can also only occur under the count-noun reading. In other words, we predict that the quantifying construction a weng a should be acceptable with the diminutive suffix, and this is indeed what we observe:

1. (9)

   

Following the data patterns above, we thus conclude that the quantifying construction a weng a crucially depends on doubling of the indefinite determiner. That construction can optionally co-occur with diminutive morphology on the noun because this morphology is subject to the same constraint of having to occur with count nouns. Based on the examples above, we propose the following basic structure of the quantifying construction a weng a, which corresponds to structural claims that have been made for Bavarian indefinite determiner doubling (Kallulli & Rothmayr 2008). In particular, we postulate that the doubling can be analyzed as a recursive DP structure, and we analyze weng as head of a Q(uantifier) Phrase:

1. (10)

   

Let us briefly summarize the East Franconian data we have introduced so far: In section 2.1, we showed that the diminutive suffix -la turns mass nouns into count nouns and thus has the same interpretive effect as numeral classifiers. Section 2.2 has introduced the quantifying construction a weng a, and we demonstrated that this linguistic device likewise turns mass nouns into count nouns and that, under the count reading, the quantifier weng can only occur with doubling of the indefinite determiner. Based on these distributional patterns, we now turn to the questions that motivated our comparison of these two linguistic devices and thus our experimental investigations of diminutive strategies in the East Franconian noun phrase.

2.3. Functions of East Franconian Diminutive Strategies: Two Hypotheses

Given the observations in sections 2.1 and 2.2, we now move on to explore a question already hinted at above: Do these East Franconian elements have a function that goes beyond their conceptualizing role of turning mass into count nouns? We submit that there are at least two such potential functions that suggest themselves. First, following the classical analysis of diminutives as a device that “means at least ‘small’” (Jurafsky Reference Jurafsky1996:534, and table 1; Schneider Reference Schneider2013), both the diminutive suffix -la and the quantifying construction a weng a could additionally be interpreted as signals that the size of the noun referent is diminished. If this is true, then they function as modifiers in the conceptual domain of magnitude. Note that intuitively this function is always associated with the diminutive suffix, but it is unclear whether or not East Franconian diminutives fulfill this function: The sheer frequency of diminutives in this variety of German may correspond to semantic bleaching (on Upper German diminutives in particular, see Kolmer Reference Kolmer1999, and earlier Schiepek Reference Schiepek1908).

Table 1. Experiment 1: Regression table for categorizations (n.s. means “not significant”).

A second way in which both the suffix -la and the construction a weng a could have a diminishing function is not within the domain of magnitude (size, volume, length, etc.), but within the domain of degrees of gradable noun phrases. That is, in addition to testing whether or not these diminutive devices play a role in measuring magnitudes of objects, we also consider the possibility that they could function in the domain of measuring degrees of predicates. This hypothesis is driven by the observation in the literature that examples such as 11 have, in fact, two interpretations (example from Kolmer Reference Kolmer1999:34).

1. (11)

   

First, 11 can refer to a small portion of bad beer; second, it can refer to the small degree to which the substance beer has gone bad. Accordingly, we observe that a weng a can also modify the degree of predicates such as the adjective schlechts ‘bad’ in 11. In order to enable the comparison between the diminutive morpheme -la and a weng a (based on their common syntactic distribution with count nouns), we focus on gradable nouns instead, and not on adjective-noun combinations. Consider the following examples, showing that some nouns denote gradable properties, as in 12a, while others do not, as in 12b (see Morzycki Reference Morzycki2009 for extensive discussion of this difference).

1. (12)

   

Since the noun idiot in 12a denotes a gradable predicate and thus has the degree reading ‘someone who is idiotic’, its combination with an adjective such as big most likely results in the reading that someone is idiotic to a high degree. In other words, the adjective does not measure the physical size of someone who is an idiot, but rather modifies the degree component of the gradable noun idiot. In contrast, nouns such as boy in 12b do not have such a degree reading and thus their combination with big most likely yields the reading ‘someone who is a boy and who is big for a boy’. Accordingly, nouns such as idiot may be thought of as containing an abstract degree head POS, which distinguishes them from nongradable cases and heads the Deg(gree) Phrase (see Kennedy & McNally 2005 on POS and Morzycki Reference Morzycki2009:188 on the following structural claim, and Morzycki Reference Morzycki2016:157–162 on Deg Phrases more generally). In 13, e is the type of entities/individuals, t is the type of truth values, and d refers to degree.

1. (13)

   

Semantically, we would expect a weng a to have an effect on degree modification, because of weng’s usual function as a mass noun modifier, indicating the quantity of a substance, as in (8). To see this, consider the conceptual parallels between the measurement of mass nouns and predicates denoting degrees. For instance, if a mass, such as ‘beer’, is divided, it stays beer; if a degree predicate such as being an idiot is reduced, the person still stays an idiot—whereas if one divides a count object such as ‘cup’, the resulting parts are not still ‘a cup’. This has been called the principle of divisiveness in the literature (Bale & Barner 2009; Wellwood Reference Wellwood2015; Wittenberg & Levy 2017).

We suggest that in order to explore the conceptual effects of diminutive strategies, one therefore should consider also this degree component of gradable nouns, in addition to their effect on measuring (physical) sizes. Accordingly, complementing our study on magnitude effects described in section 3 below, we also conducted an experiment where we tested the co-occurrence of East Franconian gradable nouns (analogous cases to idiot above; for example, Gribbl ‘crook’) with both the diminutive suffix, the quantifying construction, and a combination of the two. However, let us first start with our experiment on potential magnitude effects.

3.1. Participants, Methods, and Materials

We recruited 129 participants from Franconia through social media, through an interview of the first author to a local newspaper in Upper Franconia, and through local radio announcements. We excluded five participants because they indicated (by self-report) that they were not fluent speakers of East Franconian and/or avoided dialect whenever possible. The remaining 124 participants were on average 42 years old (range: 22–74), and 52% of them were female.

We conducted an extensive dialect questionnaire, asking about dialect use frequency, language attitude, and proficiency. All our final participants had a positive attitude toward dialect, and reported to be proficient, using dialect often and for the most part deliberately. Figure 1 shows the expected positive correlation between attitude and frequency of dialect use. For the demographics questionnaire, we used the web-based survey platform Qualtrics (Qualtrics, Provo, UT), from which we directed visitors (with their consent) to the experiment. We hosted the experiment on the servers of the University of California, San Diego, using the jsPsych (Leeuw Reference Leeuw2015) library with custom plug-ins to present this and the subsequent study and to record data.

Figure 1. Dialect use and attitude of participants in experiment 1.

Participants were presented with a cover story about having an elderly neighbor who records an audio tape in lieu of writing a shopping list when she goes to the supermarket. This was done in order to make the task more interactive and to avoid interference from orthographic forms in a dialect without a written norm. Participants were instructed to help their fictitious elderly neighbor to pick the appropriate object or objects from a shopping list. This list was recorded by the first author (a native speaker), and the audio files were spliced by research assistants such that half of the items were combined with the carrier phrase und dann breicherd ich nuch ‘and then I’d also need…’, and half of them were combined with und außerdem nuch ‘and then also…’.

There were 12 filler items varying according to number (singular versus plural) and mass versus count nouns. These served as distractor items to conceal the purpose of 16 critical items in four conditions: unmodified, as in 14a, modified by the morphological diminutive -la, as in 14b, modified by both -la and the quantifying construction a weng a, as in 14c, and modified by a weng a, as in 14d.^{Footnote 4}

1. (14)

   

The presentation of trials was Latin-squared, such that each participant never heard a given expression more than once, but heard each modification option four times, in combination with four different nouns; the order of items was pseudorandomized. Participants selected the matching items from an array of pictures by clicking on the pictures themselves. For fillers, there were 3–6 pictures to choose from, all chosen to represent distractors or close semantic neighbors of the respective filler item (see https://osf.io/n82c5/, Experiment I, for pictures and audio recordings). Critical items were the same pictures in four different sizes. The sizes were created by taking a reference picture and by reducing its size by a quarter, half, and three-quarters (see figure 2).

Figure 2. Example stimuli used in experiment 1 (filler on the left; critical item on the right).

The dependent variable was the size of the selected picture. If morphological diminutives are not bleached in East Franconian, sentences such as 14b,c should result in choices of smaller average sizes than sentences with nonmodified nouns, as in 14a. The same logic applies to the the quantifying construction a weng a.

3.2. Results and Discussion

Performance on filler items was at ceiling, with an overall accuracy rate of 94.7% (range: 77%∼100%). The only filler item receiving less than 91% accuracy was a Flaschn Wein ‘a bottle of wine’, which some participants interpreted as plural, selecting a competitor picture that displayed multiple bottles of wine. The high accuracy rates on the filler items indicate that participants were taking the task seriously and had no trouble following the instructions.

We analyzed the data using a mixed-effects cumulative logit model, implemented by R’s ordinal package (Christensen Reference Christensen2019): For ordered response categories (in our case, four bins ranging from smallest to biggest), this kind of model specifies response probabilities for a given choice of bin as a function of predictor variables—in our case, how the noun was modified. Instead of the intercept, ordered logit models provide a set of threshold parameters, which describe the boundaries from one bin to the next, and the probability of being drawn from one particular bin is estimated by the linear predictors with the inverse logit function (Wittenberg & Levy 2017). Since model comparisons (Levy Reference Levy2014) showed that none of the demographic factors explained a significant amount of the variance and including them did not change the pattern of the results, we only report the simplest model here, with only modifier as fixed predictor, and item and subject as random intercepts.

Table 1 contains an overview of the results of the cumulative logit model, which confirm this impression statistically. The first three rows indicate the threshold coefficients from one bin (smallest/small/big/ biggest) into the next. The regression coefficient (β) for each modifier is shown in the next part, with “no modification” as a baseline: For instance, the value of -.16 for the quantifying construction would mean that a weng a + NPs are associated with smaller sizes. However, none of the predictors are significant: Neither diminutive option—the quantifying construction, the morphological diminutive, or their combination—had any effect on which size of objects people chose.

Figure 3 shows, coded by color, the proportion of picture sizes chosen, depending on how the item was described—with the morphological diminutive -la, the quantifying construction a weng a, both, or neither. As can be seen at first glance, most choices are close to or at chance level.

Figure 3. Experiment 1: Proportion of picture sizes chosen, per diminutization option.

This experiment asked two questions: We investigated whether the morphological diminutive -la and the quantifying construction a weng a affect how people conceptualize the magnitude of noun referents, using a phrase-picture matching task: Participants listened to a grocery list and clicked on the best candidate items from an array of pictures. The results show that indeed, people do not pick smaller referents out of an array if the referent is described using the morphological diminutive -la. This was expected, based on how frequently morphological diminutives are reported to be used in Upper German dialects, presumably resulting in a bleached-out function (see our discussion in section 2 above). However, the results also show that the quantifying construction a weng a likewise has no effect on how people decide on a referent for a noun. In sum, people picked the referent at chance level, regardless of whether a noun was modified by -la, a weng a, or even both -la and a weng a.

However, as argued above, the function of both the morphological diminutive -la and that of the quantifying construction a weng a cannot only be analyzed in terms of magnitude, but also in terms of degree. In the second experiment, we ask whether these two ways of forming diminutives in East Franconian result in how people make degree judgments.

4.1. Participants, Methods, and Materials

We recruited 109 participants through the same channels as in experiment 1. We excluded 1 participant because they indicated (by self-report) that they never speak East Franconian. The remaining 108 participants were between 20 and 81 years old (average age: 45), and 51% of them were female. As in experiment 1, our participants’ language attitude was quite positive, and their usage for the most part frequent. Figure 4 shows the participants’ responses with respect to their dialect and language attitude. The heatmap (top) indicates the participants’ dominant dialect. The bottom figure correlates participants’ self-assessed frequency of dialect use and their willingness to use it if they were to be interviewed on national news (“Tagesschau”) as an indication of language attitude. To host this experiment, including the demographics questionnaire, we only used the web-based survey platform Qualtrics (Qualtrics, Provo, UT).

Figure 4. Participants’ dominant dialect and attitude.

Participants were instructed to rate the degree to which a predicate holds in a gradable noun; that is, they assessed the “strength” of modified or unmodified nouns on a scale from 1 to 10. As in experiment 1, we created eight critical trials (as in 15a–d), and eight filler trials (for a full list of stimuli, see Supplemental Material). Both the nouns and the adjectives used in the evaluation task were taken from a local dialect dictionary (Müller 2013). As discussed in section 2 above, gradable nouns are often those that have clear evaluative concepts associated with them (see the prominent example idiot in Morzycki Reference Morzycki2009); hence, we used mild invectives, such as Gribbl ‘crook’, Schlawiner ‘mischievous person’, or Fregger ‘mischievous child’ (for a full list of stimuli, see https://osf.io/n82c5/, Experiment II).

1. (15)

   

In addition to the eight critical items, we also used eight fillers. Those were also gradable nouns (that is, Schnorrer ‘freeloader’, Dadderer ‘very old man’). These were presented with a range of adjectival or adverbial modifications, but never with the modifiers under investigation. In three of the fillers, we asked for typical properties (that is, how “gesprächig” ‘loquacious’ a “rechda Ladschkabbm” ‘a.pretty talkative.person’ was). We expected relatively strong (50% and above) judgments for these fillers.

Since we expected stronger judgments for critical items as well, we countered the expected imbalance by also including five fillers in which we asked for the polar opposite of the scalar predicate (that is, how “freundlich” ‘friendly’ a “ganz a Oarlicher” ‘a.very grumpy.person’ was). As in experiment 1, the presentation of trials was Latin-squared, such that each participant never saw a given expression more than once. Our dependent variable was the evaluation strength of adjectival descriptions (that is, gerissen ‘cunning’). An example of a trial is shown in figure 5.

Figure 5. Evaluation task in Experiment II (unmodified condition).

4.2. Results and Discussion

The fillers were rated as expected: Nouns whose typical property was evaluated received average ratings between 70.7% and 88.9% (range: 4%∼100%); nouns, for which the polar opposite of their typical property was evaluated received average ratings between 10.3% and 31.8% (range: 0%∼100%). This again is an indication that participants understood the task, were paying attention, and used the full range of the scale.

The mean rating per critical condition and the standard error are shown in figure 6. The overall rating patterns were analyzed using a binomial mixed effects regression on dummy-coded data (Baayen et al. Reference Baayen, Davidson and Bates2008), implemented with R’s lme4 package (Bates et al. Reference Bates, Martin, Ben, Steven, Rune Haubo Bojesen, Henrik, Dai, Fabian and Rcpp2014). We coded the unmodified expression as a baseline condition (no modification, the quantifying construction a weng a, the morphological diminutive -la, or both) and location (coded as being from Hof/Rehau versus all other areas) as fixed effects, and random slopes of condition and age of participant, item, and participant (Barr et al. Reference Barr, Roger, Christoph and Tily2013). We also conducted planned pairwise comparisons with contrast-coded data in a binomial model using the same effects structure as for the omnibus analysis, collapsing across conditions, and Bonferroni-correcting for multiple comparisons.

Figure 6. Mean strength ratings of gradable nouns (with SEs) in experiment 2.

As table 2 shows, Location had a significant effect on ratings: Speakers from the Hof/Rehau region of Franconia rated the expressions as stronger than people outside that region (β = -0.89, p<.001). Crucially, as in experiment 1, the morphological diminutive (Schlawiner-la, 79.9%) did not result in ratings different from the unmodified expression (Schlawiner, 81.8%). However, both the quantifying construction (a weng a Schlawiner, 76%) alone, and in combination with the morpho-logical diminutive (a weng a Schlawiner-la, 74.9%), resulted in significantly lower ratings than the unmodified expression.

Table 2. Results of regression analyses, experiment 2 (top: omnibus effects, bottom: planned comparisons.) Significant effects are in bold.

For both pairwise analyses, Location was again a strong predictor. More importantly, the first pairwise analysis compared ratings for conditions containing the morphological diminutive (see 15c,d) to the conditions not containing it (see 15a,b). Results reported in the bottom half of table 2 suggest that the presence or absence of the morphological diminutive made no difference for ratings. The second pairwise analysis compared ratings for conditions containing the quantifying construction (see 15b,d) to the conditions not containing it (see 15a,c), and the difference was highly significant. These results indicate that the presence or absence of the morphological diminutive had no effect on how strongly people rated the degree to which a predicate holds in gradable nouns but that the addition of the quantifying construction a weng a reduced the degree of the same predicate in gradable nouns significantly.

This study investigated the effect of the morphological diminutive -la and the quantifying construction a weng a on the degree semantics of gradable nouns, such as Gribbl ‘crook’. Participants were instructed to rate the degree to which a predicate holds in gradable modified or unmodified nouns, such as Gribbl ‘crook’. Again, the morphological diminutive -la did not significantly change the degree to which a predicate holds in a gradable noun, but we found that a weng a did. We also found a significant effect of location: Dialect speakers from the region around Rehau rate the degrees of the predicates of the gradable nouns stronger than everyone else. This is an interesting effect of dialectal microvariation onto the conceptual effects of diminutization. However, one cannot exclude two other factors that may explain this finding: First, the vast majority of our participants were from this corner of Upper Franconia, and it might be that the relative scarcity of data from other regions of Upper Franconia in the statistical model skewed the weight of this group’s result. Second, this may be a familiarity effect, since both the gradable nouns and the adjectives were taken from a region-specific dialect lexicon (Müller 2013): The perceived intensity of a gradable noun may be stronger when one is more familiar with it. While the first factor is hard to control for in large-scale, citizen science experiments, the second possibility could be explored by using expressions from several dialect dictionaries, casting a wider net across the geographic distribution of lexical expressions.