28 March 2025
Jiyeong Kim: Diachronic Semantic Changes of meri ‘head’ and the Classifier mari in Korean
Human body terms actively serve as a source domain for the conceptualization of entities beyond the human body and often undergo grammaticalization (Svorou 1993; Heine 1997; Kraska-Szlenk 2014). In Modern Korean, meri and mari function distinctly as a body part term and a classifier for counting animals, respectively. However, historical evidence indicates that both meri and mari originally referred to the body part ‘head.’ This study examines the diachronic semantic changes of meri and mari using the Korean History Corpus (2024), providing a detailed analysis of how these changes are reflected in corpus data. While both terms denoted the body part ‘head’ from the 15th century, their meanings changed through three stages: semantic divergence (15th–16th century), intermediate convergence (17th–18th century), and re-divergence (19th century–present). Furthermore, it was shown that the use of meri ‘head,’ following the metaphoric chain of ‘person > object > space’ in the 15th century, exhibits a further directional extension to ‘person > object > space > time’ by the 19th century.
Olabode Adedeji: Bare Nominals and Definiteness in Yorùbá
Yorùbá, a Kwa language, has a definite determiner, náà, and allows for definite interpretations of bare nouns. Theoretically, languages should eliminate covert type-shifting operations, provided there is a lexical exponent with the same meaning: the Blocking Principle (Chierchia, 1998). Therefore, a language like Yorùbá, which accommodates both lexical and covert derivation of the same meaning, presents an interesting case study. Adopting the neo-Carlsonian theory of cross-linguistic variation (Chierchia, 1998; Dayal, 2004), this study explores the semantic nuances of bare nouns and náà-marked definites in Yorùbá. The study shows that while covert definite iota and náà are available for definiteness, they do not convey the same meaning and are not semantically exchangeable. Following Dayal & Jiang (2020) and Owusu (2021), we illustrate that náà encodes a non-uniqueness presupposition that covert iota does not.
4 April 2025
María Teresa Borneo: The pervasiveness of metonymy over time: the case of English Manner of Motion verbs
This project investigates the diachronic trajectory of nine English Manner of Motion (MoM) verbs’ metonymic extension, such as the one in “I bounced ideas with my coworkers”. Through a corpus study, the goal is to discern between two explanations of metonymy as a meaning change mechanism. On the one hand, the Diachronic Metonymization hypothesis, which aligns with a literalist notion in which each lexical item has an original, denotational meaning, independent of context of usage, that gets manipulated in different ways to convey a ‘figurative’ meaning (Grice, 1975; Sperber and Wilson, 1986; Katz, 1980; Ariel, 2002). This hypothesis predicts that MoM verbs become more metonymic with time, reflected in their frequency distributions. On the other hand, the Pervasiveness of Metonymy hypothesis postulates that metonymy is a meaning association phenomenon intrinsic to the conceptual system, and as such, it does not occur as a later development or an evolving trend but is rather constant. Under this approach, we propose that metonymy operates through a deconstraining of selectional restrictions, instead of a change in lexical structure, which allows for the highlighting of the relevant aspects (manner). This predicts that frequency of metonymic senses will be equal for all verbs, since they share the same structure, and metonymic usages will be present since very early attestations.
Results show that, even if metonymic senses are less frequent than non-metonymic, the prediction that metonymic frequencies increase over time does not bear out and neither does the hypothesis that all verbs would show the same distribution. Instead, we observe that metonymic senses of these verbs are documented from very early attestations, giving support to the Pervasiveness hypothesis, and that each verb has a different metonymic frequency distribution. Moreover, by using semantic vectors (Global Vectors for Word Representation [GloVe]) and clustering the nouns that co-occur with these verbs, we observe a high preference of speakers to extend metonymically using nouns that refer to human body parts more than any other semantic domain, a difference that is stable across time periods. This relates to a speakers’ tendency to privilege the parts of the lexicon which they can more closely relate to.
Dani Katenkamp: Lack of prestige in use of English borrowings in nineteenth century Choctaw society
The European colonization of the Americas and later industrialization created an environment in which Native communities needed to create a large number of new ‘words,’ i.e. neologisms, to describe the new institutions and phenomena in their environment. Communities met this need with a variety of neologizing strategies, only some of which involved borrowing from European languages. In this presentation I examine neologizing behavior in nineteenth century Choctaw as a case study.
My empirical goal is to elaborate our understanding of how different Choctaw social classes interpreted the relationship between the administration of Choctaw Nation and the culture of the United States. Theoretically, I show that in addition to community-wide attitudinal features regarding multi-community interaction, there are nuanced attitudinal features of subgroups within a community that influence borrowing behavior. Additionally, I will argue that while speakers may simply feel positively or negatively about the practice of borrowing, they may also think of borrowed terms in relation to potential social prestige, cultural shift, or other group-internal social issues.
11 April 2025
Finn Amber: Switch Reference between coordinated clauses in Historical Choctaw: Implications for Agree
In this talk, I also propose a new analysis of Switch Reference (SR) in Historical Choctaw, arguing that this data motivates an extension to Agree which allows agreement between constituents that are not, initially at least, in a c-command relation. Using data from the Historical Choctaw Corpus (Katenkamp, Broadwell, & Wolf, 2020), I propose that SR occurs between coordinated clauses in Historical Choctaw; a possibility that has been the subject of significant debate in the SR literature (see eg. Broadwell, 1997 vs McKenzie, 2015). I demonstrate that an Agree-based analysis (eg. Baker & Souza, 2019) of SR is preferable over a binding-based one (e.g. Finer, 1985). Following this, I show that SR requires agreement between the SR marker and the subject of the higher clause, syntactic units that are not in a c-command relation. Doing this straightforwardly violates basic locality considerations which require that there be a c-command relation between Agreeing elements (Chomsky, 2000). Therefore, in order to account for SR without such a serious theoretical compromise, there must be some extension to the range of Agree, along the lines of Aboutness Topics (Frascarelli, 2007), or Cyclic Agree (Rezac, 2003).
Miranda Zhu: The Structural Sources of Verb Meaning in Large Language Models
Experiments in language acquisition have shown that human learners perform syntactic bootstrapping: they use syntactic information to acquire verb meanings. In this talk, we explore whether language models like ChatGPT exhibit syntactic bootstrapping as well. We train models under two different transformer architectures, BERT and GPT-2, on perturbed child-directed speech where either co-occurrence information or syntactic information was removed. Results show that removing syntactic information during training negatively influences the learned representation of verbs. Moreover, for BERT, removing co-occurrence information has a larger negative influence on the representations of physical verbs (like push or play) as compared to mental verbs (like think or like), which is consistent with experimental findings in humans. Overall, these results allow us to better interpret the learning mechanism of transformer language models and provide further support for the important role of syntactic structure as a source of information about verb meaning during language acquisition.
18 April 2025
Alessandra Pintado-Urbanc: Reexamining Mass/Count Flexibility in the Nominal Domain
Are the mass-to-count or ‘portioning’ uses (‘several beers’) and count-to-mass or ‘grinding’ uses (‘a bit of pear’) the result of lexical derivations with real-time processing effects? The evidence is inconclusive. While Frisson & Frazier (2005) argue that these semantic derivations are visible as cost, Lima (2019) reports no such effect exists. In this presentation, I address this disparity through two English self-paced reading experiments using Frisson & Frazier (2005)’s materials. Experiment I tested the ‘portioning’ rule (‘several pears’ vs. ‘several beers’). No cost was found at the word nor sentence level. Experiment II tested the ‘grinding’ rule (‘a bit of beer’ vs. ‘a bit of pear’). Results show higher reading times for the ‘derived’ conditions one word after the critical noun. Moreover, a historical corpus analysis was also conducted to examine the diachronic trajectories of these derivations over a 200 year time period. Once again, there is an asymmetry with mass-to-count derivations increasing over time and count-to-mass derivations decreasing over time. These results, along with findings from acquisition, object processing and adult behavioral data, provide empirical support for an asymmetrical relationship between mass and count nouns thus motivating a new understanding of the conceptualization of the mass/count distinction.
Richard Luo: I wonder if Georgian can tell us how tu embed(s) questions?
What is the underlying structure of a question, and how is its meaning generated by the grammar? This talk will examine the interactions between constituents of interrogative clauses at the syntax-semantics interface, focusing on the different ways in which wh-questions and polar questions are built up compositionally in Georgian (Kartvelian). Special attention will be paid to the interrogative particle tu found in embedding contexts; it will be argued that tu is an interrogative complementizer, but differs from English whether in certain crucial respects – (i) it can occur optionally in embedded wh-clauses, in which case it must immediately precede the fronted wh-phrase, and (ii) it cannot occur in embedded questions with alternative disjunction tu ara ‘or not.’ I will present an analysis of tu which accounts for both of these distributional properties, departing from the standard treatment of whether attributed to Hamblin-Karttunen question semantics. It does not establish interrogative clause-typing, but assumes its argument to already have the semantic type of a question; instead, what tu contributes is a non-singleton presuppositional constraint.
Besides alternative disjunction tu ara ‘or not,’ there is a second strategy for polar question embedding which uses only tu and places a strict requirement on the surface word order. Such contexts, I will argue, require tu to appear in the immediate preverbal focus position of the embedded clause, resulting in displacement of other non-focal material into the clausal periphery to satisfy prosodic alignment with narrow focus (Borise, 2023). Semantically, the derivation would crash due to a presupposition failure with tu, but it may be pragmatically repaired via Anti-Singleton Coercion (Biezma & Rawlins, 2012). I will show that this coercion operation is constrained and cannot apply universally, predicting certain environments with subordinate interrogative clauses to not permit embedding polar questions with tu. Overall, the paradigm reveals Georgian embedded interrogatives to require a constituent in focus position, reinforcing the tight association between questions and focus (Hagstrom, 1998; Kratzer and Shimoyama, 2002; Beck 2006; i.a.).
25 April 2025
Squid Tamar-Mattis: Auxiliary agreement in Basque: evidence from backward gapping
Basque finite auxiliaries mark the person and number of their ergative, absolutive, and dative arguments. The prevailing view is that at least some of these markers are clitics, though there is some disagreement about exactly which ones. I present data like (1) which shows that Markina Basque permits summative agreement, creating a problem for clitic-doubling analyses.
(1) [Jon-ek ure] eta [ni-k teie] edaten do-gu-z.
Jon-ERG water.ABS and 1SG-ERG tea.ABS drink.IMPF AUX-1PL.E-PL.A
‘I drink tea and Jon water.’
Here the auxiliary seems to get the 1PL ergative marker -gu by “combining” (in some sense) the ergative arguments from two separate clauses, and the same thing seems to happen with the absolutive arguments. However, if we are inclined to view -gu as a 1PL clitic, there is no 1PL DP in which it could originate. The same problem can be demonstrated for absolutive and dative arguments. Building on a framework developed by Grosz (2015), I show that this problem can be solved under an agreement-based analysis. I also discuss several possible ways to repair the clitic-doubling analysis and show that they either fail to explain the data or lead to larger theoretical problems.