Issue of Monday, March 6th, 2023
Exp/Comp Group 3/10: Hayley Ross (Harvard)
Date/Time: Friday 3/10 from 2-3:30pm
Location: 32-D831 and on Zoom
Speaker: Hayley Ross (Harvard)
Title: Adjective-noun compositionality in humans and language models
Abstract: A key component of human language is compositionality: the idea that we assemble the meaning of a sentence or phrase in a structured process from the meanings of its parts. When language models encounter a phrase like blue cup or fake gun, do they engage in any compositional process? Do they know that a blue cup is a thing which is blue and which is a cup – but that a fake gun is not a gun? I focus on adjective-noun compositionality, specifically with so-called privative adjectives, to try to get insight into whether neural language models learn about compositionality from their next word prediction objective, or whether they are merely memorising specific phrases. For that matter, do humans treat rare but plausible phrases like counterfeit scarf compositionally, or do they actually also rely on conventionalised meaning to know how to interpret such combinations? What is the relationship between frequency and ease of understanding? In this talk, I’ll present a series of experiments designed to tackle these questions.
LingLunch 3/9 - Keny Chatain (Institut Jean Nicod (ENS))
Speaker: Keny Chatain (Institut Jean Nicod (ENS))
Title: Reducing Pronoun Accessibility To Presupposition Satisfaction
Time: Thursday, March 9th, 12:30pm - 2pm
Abstract: A pronoun cannot always co-refer with a given DP: while the cases in (1a-c) are natural, the cases in (1d-e) are not interpretable. The problem of pronoun accessibility is the problem of determining which antecedent-pronoun configuration are licit, which ones give rise to deviance. The cases in (1) point to a simple generalization: the pronoun “it” can co-refer with “a phone-book” if and only if the existence of a phone-book can be taken for granted at the point where the pronoun “it” is used. Simple though it may seem, this generalization is not fully validated by many current theories of pronouns. Such theories typically under-generate, failing to license cases like (1c) as well as other more complicated examples.
(1)
a. There is a phone-book7 and it7 is in the cabinet.
b. If there is a phone-book7, it7 is in the cabinet.
c. Either there isn’t a phone-book7 and it7 is in the cabinet. (attributed to Partee)
d. # Either there is a phone-book7 or it7 is in the cabinet.
e. # There might be a phone-book7 and it7 is in the cabinet.
Taking the generalization at face-value, I propose a system where a pronoun can be interpreted if and only if the existence of a witness - a phone-book in the cases above - can be presupposed. This theory builds on insights from E-type theories (Evans, 1980 ; Heim, 1990 ; Elbourne, 2005) but drops some of the assumptions that have made such theories inviable, like uniqueness.
The benefits are conceptual and empirical. Conceptually, by reducing pronoun accessibility to presupposition satisfaction, the proposal can build upon so-called “explanatory” theories of presuppositions (Schlenker, 2009 ; George, 2008 ; Fox, 2013, a.o.). Such theories derive discourse effects from truth-conditional meaning, instead of baking these effects into meanings themselves (Soames, 1989), as in Dynamic Semantics (Heim, 1983, a. o.). Second, the proposal has a broad empirical coverage: it explains the original cases in (1), but also the more complex quantified cases in (2) of quantifier subordination (Roberts, 1987) and donkey anaphora (Geach, 1962). In addition, it makes a range of new predictions, regarding the possibility of cataphora and pronouns licensed by pragmatic inferences.
(2)
a. Every farmer who has a donkey feeds it hay.
b. Every farmer has a donkey. Few of them feed it hay.
Speaker: Keny Chatain (Institut Jean Nicod (ENS))
Title: Reducing Pronoun Accessibility To Presupposition Satisfaction
Time: Thursday, March 9th, 12:30pm - 2pm
Abstract: A pronoun cannot always co-refer with a given DP: while the cases in (1a-c) are natural, the cases in (1d-e) are not interpretable. The problem of pronoun accessibility is the problem of determining which antecedent-pronoun configuration are licit, which ones give rise to deviance. The cases in (1) point to a simple generalization: the pronoun “it” can co-refer with “a phone-book” if and only if the existence of a phone-book can be taken for granted at the point where the pronoun “it” is used. Simple though it may seem, this generalization is not fully validated by many current theories of pronouns. Such theories typically under-generate, failing to license cases like (1c) as well as other more complicated examples.
(1)
a. There is a phone-book7 and it7 is in the cabinet.
b. If there is a phone-book7, it7 is in the cabinet.
c. Either there isn’t a phone-book7 and it7 is in the cabinet. (attributed to Partee)
d. # Either there is a phone-book7 or it7 is in the cabinet.
e. # There might be a phone-book7 and it7 is in the cabinet.
Taking the generalization at face-value, I propose a system where a pronoun can be interpreted if and only if the existence of a witness - a phone-book in the cases above - can be presupposed. This theory builds on insights from E-type theories (Evans, 1980 ; Heim, 1990 ; Elbourne, 2005) but drops some of the assumptions that have made such theories inviable, like uniqueness.
The benefits are conceptual and empirical. Conceptually, by reducing pronoun accessibility to presupposition satisfaction, the proposal can build upon so-called “explanatory” theories of presuppositions (Schlenker, 2009 ; George, 2008 ; Fox, 2013, a.o.). Such theories derive discourse effects from truth-conditional meaning, instead of baking these effects into meanings themselves (Soames, 1989), as in Dynamic Semantics (Heim, 1983, a. o.). Second, the proposal has a broad empirical coverage: it explains the original cases in (1), but also the more complex quantified cases in (2) of quantifier subordination (Roberts, 1987) and donkey anaphora (Geach, 1962). In addition, it makes a range of new predictions, regarding the possibility of cataphora and pronouns licensed by pragmatic inferences.
(2)
a. Every farmer who has a donkey feeds it hay.
b. Every farmer has a donkey. Few of them feed it hay.
Colloquium 3/10 - Vera Gribanova (Stanford University)
Speaker: Vera Gribanova (Stanford University)
Title: On the sources of case (dis-)connectivity in two types of Russian TP ellipsis
Time: Friday, March 10th, 3:30pm - 5pm
Abstract: Discussions of the identity relation in constituent ellipsis licensing often take for granted, either explicitly or implicitly, the idea that the identity relation in ellipsis ought to be uniform, and applicable across different ellipsis configurations and languages. Recent investigations of this relation — Rudin 2019, Kroll 2019, Anand, Hardt, and McCloskey 2022, Stigliano 2022 — have provided novel evidence and arguments in support of the view that the domain of application of the identity relation is not always coextensive with the domain of the ellipsis itself. For example, although the prevailing view of English sluicing as TP ellipsis historically took the domain of the identity relation to likewise be the TP (Merchant 2001), one of the main findings of the UCSC sluicing dataset is that material above the level of vP — e.g. tense/finiteness, modality, and polarity — can undergo felicitous mismatches. Generalizing beyond English sluicing, this raises the question of whether the domain of ellipsis identity must be a proper subset of the domain of ellipsis itself, or if the specific size of the domain relevant for the identity relation may be variable across languages and ellipsis configurations.
In this talk, I present an investigation of some asymmetries in how case connectivity is enforced in two types of Russian clausal (TP) ellipsis — contrastive polarity ellipsis and fragment answers — and develop an analysis explaining why these asymmetries take the shape that they do. The case study leverages the availability of a well-known case alternation between structural (nominative/accusative) case and genitive case under negation. The first asymmetry is that case connectivity on remnants of these two ellipsis types is enforced fully only in fragment answers, but not in contrastive polarity ellipsis, in which a contrastive DP is fronted to the left periphery, preceding a polar particle (‘yes’ or ‘no’). The second asymmetry is that in contrastive polarity ellipsis, genitive patients under negation in the antecedent can correspond to an accusative patient remnant outside the ellipsis site, but not the reverse. To capture these asymmetries, I develop an analysis of the system of licensing relations that connects the syntax of polarity expression, negative concord, and genitive of negation, and combine this with a formulation of the identity relation in ellipsis in which head-to-head identity between the elided material and the antecedent must be invoked (Saab 2008, 2010, 2022, Tanaka 2011, Rudin 2019, Stigliano 2022).
For the asymmetries between these two types of Russian TP ellipsis to emerge within an internally consistent system of analytical commitments, it is critical that the domain of evaluation for identity be larger than in English sluicing, and likely coextensive with the elided TP. This finding supports a view in which the domain of evaluation for the identity relation in ellipsis may vary across languages and ellipsis types. In the latter part of the talk, I point to a view of ellipsis licensing that can straightforwardly capture such variation, and which arises directly from unifying existing analyses (Aelbrecht 2010, Stigliano 2022) in which certain sub-parts of the ellipsis function — non-pronunciation, syntactic licensing, and the identity relation — can be either grouped together, or broken up across several distinct heads in the clausal spine.
References
Aelbrecht, Lobke. 2010. The syntactic licensing of ellipsis. John Benjamins.
Anand, Pranav, Daniel Hardt, and James McCloskey. To appear. The domain of formal matching in sluicing. Linguistic Inquiry.
Kroll. Margaret. 2019. Polarity reversals under sluicing. Semantics and Pragmatics 12.
Merchant, Jason. 2001. The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis. Oxford: Oxford University Press.
Ross, John Robert. 1969. Guess who? In Proceedings of the 5th Annual Meeting of the Chicago Linguistic Society. Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan (eds.), 252–286.
Rudin, Deniz. 2019. Head-based syntactic identity in sluicing. Linguistic Inquiry 50 (2): 253–283.
Tanaka, Hidekazu. 2011. Syntactic identity and ellipsis. The Linguistic Review 28: 79–110.
Saab, Andrés. 2008. Hacia una teoría de la identidad parcial en las elipsis. PhD Dissertation, Universidad de Buenos Aires.
Saab, Andrés. 2010. Silent interactions: Spanish TP-ellipsis and the theory of island repair. Probus 22 (1): 73–116.
Saab, Andrés. 2022. Grammatical silences from syntax to morphology. A model for the timing of ellipsis. In The Derivational Timing of Ellipsis, ed. Güliz Günes and Anikó Lipták. Oxford: Oxford University Press.
Stigliano, Laura. 2022. The silence of syntax: A theory of ellipsis licensing and identity. PhD dissertation, University of Chicago.
Speaker: Vera Gribanova (Stanford University)
Title: On the sources of case (dis-)connectivity in two types of Russian TP ellipsis
Time: Friday, March 10th, 3:30pm - 5pm
Abstract: Discussions of the identity relation in constituent ellipsis licensing often take for granted, either explicitly or implicitly, the idea that the identity relation in ellipsis ought to be uniform, and applicable across different ellipsis configurations and languages. Recent investigations of this relation — Rudin 2019, Kroll 2019, Anand, Hardt, and McCloskey 2022, Stigliano 2022 — have provided novel evidence and arguments in support of the view that the domain of application of the identity relation is not always coextensive with the domain of the ellipsis itself. For example, although the prevailing view of English sluicing as TP ellipsis historically took the domain of the identity relation to likewise be the TP (Merchant 2001), one of the main findings of the UCSC sluicing dataset is that material above the level of vP — e.g. tense/finiteness, modality, and polarity — can undergo felicitous mismatches. Generalizing beyond English sluicing, this raises the question of whether the domain of ellipsis identity must be a proper subset of the domain of ellipsis itself, or if the specific size of the domain relevant for the identity relation may be variable across languages and ellipsis configurations.
In this talk, I present an investigation of some asymmetries in how case connectivity is enforced in two types of Russian clausal (TP) ellipsis — contrastive polarity ellipsis and fragment answers — and develop an analysis explaining why these asymmetries take the shape that they do. The case study leverages the availability of a well-known case alternation between structural (nominative/accusative) case and genitive case under negation. The first asymmetry is that case connectivity on remnants of these two ellipsis types is enforced fully only in fragment answers, but not in contrastive polarity ellipsis, in which a contrastive DP is fronted to the left periphery, preceding a polar particle (‘yes’ or ‘no’). The second asymmetry is that in contrastive polarity ellipsis, genitive patients under negation in the antecedent can correspond to an accusative patient remnant outside the ellipsis site, but not the reverse. To capture these asymmetries, I develop an analysis of the system of licensing relations that connects the syntax of polarity expression, negative concord, and genitive of negation, and combine this with a formulation of the identity relation in ellipsis in which head-to-head identity between the elided material and the antecedent must be invoked (Saab 2008, 2010, 2022, Tanaka 2011, Rudin 2019, Stigliano 2022).
For the asymmetries between these two types of Russian TP ellipsis to emerge within an internally consistent system of analytical commitments, it is critical that the domain of evaluation for identity be larger than in English sluicing, and likely coextensive with the elided TP. This finding supports a view in which the domain of evaluation for the identity relation in ellipsis may vary across languages and ellipsis types. In the latter part of the talk, I point to a view of ellipsis licensing that can straightforwardly capture such variation, and which arises directly from unifying existing analyses (Aelbrecht 2010, Stigliano 2022) in which certain sub-parts of the ellipsis function — non-pronunciation, syntactic licensing, and the identity relation — can be either grouped together, or broken up across several distinct heads in the clausal spine.
References
Aelbrecht, Lobke. 2010. The syntactic licensing of ellipsis. John Benjamins. Anand, Pranav, Daniel Hardt, and James McCloskey. To appear. The domain of formal matching in sluicing. Linguistic Inquiry. Kroll. Margaret. 2019. Polarity reversals under sluicing. Semantics and Pragmatics 12. Merchant, Jason. 2001. The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis. Oxford: Oxford University Press. Ross, John Robert. 1969. Guess who? In Proceedings of the 5th Annual Meeting of the Chicago Linguistic Society. Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan (eds.), 252–286. Rudin, Deniz. 2019. Head-based syntactic identity in sluicing. Linguistic Inquiry 50 (2): 253–283. Tanaka, Hidekazu. 2011. Syntactic identity and ellipsis. The Linguistic Review 28: 79–110. Saab, Andrés. 2008. Hacia una teoría de la identidad parcial en las elipsis. PhD Dissertation, Universidad de Buenos Aires. Saab, Andrés. 2010. Silent interactions: Spanish TP-ellipsis and the theory of island repair. Probus 22 (1): 73–116. Saab, Andrés. 2022. Grammatical silences from syntax to morphology. A model for the timing of ellipsis. In The Derivational Timing of Ellipsis, ed. Güliz Günes and Anikó Lipták. Oxford: Oxford University Press. Stigliano, Laura. 2022. The silence of syntax: A theory of ellipsis licensing and identity. PhD dissertation, University of Chicago.
Bergül Soykan @ Tu+ 8
The annual Workshop on Turkic and Langugages in Contact with Turkic (TU+8) was held at Harvard University on March 4-5. Our first-year student Bergül Soykan gave a talk titled “The interaction between past and conditional morphemes in Turkish.” You may find the abstract here.
The annual Workshop on Turkic and Langugages in Contact with Turkic (TU+8) was held at Harvard University on March 4-5. Our first-year student Bergül Soykan gave a talk titled “The interaction between past and conditional morphemes in Turkish.” You may find the abstract here.