Franklin Institute Symposium
The Past,
Present and Future of Formal Semantics
Featuring Barbara Partee
Recipient of the 2021
Benjamin Franklin Medal
in Computer and Cognitive
Science
Sponsors
The Franklin Institute
Awards Program
Department of Linguistics
Department of Computer and Information Science
Initiative in Integrated Language Science and Technology
MindCORE
University of Pennsylvania
Monday April 19th, 2021
Link to recording
9:45am (Eastern Daylight Time): Opening
10:00-10:40am: Barbara Partee (UMass Amherst)
Personal notes on the history of formal semantics [Slides]
I’ve given many talks on the history of formal semantics while I’ve been working on a book on the topic over the past ten
years. This time I want to share personal notes that illustrate the mysterious combination of serendipity and apparent inevitability that mark the emergence and development of this new field, highlighting the important role of human interactions.
Personal notes on the history of formal semantics [Slides]
I’ve given many talks on the history of formal semantics while I’ve been working on a book on the topic over the past ten
years. This time I want to share personal notes that illustrate the mysterious combination of serendipity and apparent inevitability that mark the emergence and development of this new field, highlighting the important role of human interactions.
10:40am-11:20am:
Gennaro Chierchia (Harvard University)
Language and Natural Logic: Ways in which grammar empowers our spontaneous logicality [Slides]
Gennaro Chierchia (Harvard University)
Language and Natural Logic: Ways in which grammar empowers our spontaneous logicality [Slides]
The classical division of labor between syntax and semantics is along the following lines: syntax determines which configurations of symbols are well formed; semantics maps well formed structures into their denotations / truth conditions and thereby determines what entails what, what presupposes what, etc. Something that has come up over and over in the course of linguistic investigations in the past half a century is that sentences that we want to regard as ‘ungrammatical’, including, e.g. NPI violations of the form *I have ever eaten those cookies, owe their status not to syntactic ill-formedness but the fact that they turn out to be either contradictory or tautologous, i.e. Logically-determined. This immediately raises the question of why not all L-determined sentences are felt as ‘not being in the language’, and more generally it forces us to re-think many of Partee’s key concerns, like the relation between language and logic, the role of truth-conditions and model-theory, type-shifting, etc..
11:20am-noon: Pauline Jacobson (Brown University)
Channeling Montague, Variable Free Logic(s) and more: A small puzzle with large implications [Handout]
This talk will focus on three leading ideas: (a) Direct compositionality (syntax and semantics work in tandem to prove expressions well-formed and assign them a meaning (no notion of ‘levels’ and no ‘logical form’ for interpretation); (b) the tight ‘fit’ between syntax and semantics afforded by a Categorial Grammar; (c) a semantics making no use of variables. (a) and (b) come directly from ideas of Montague and used in ‘classical’ Montague grammar, whose development, importance and applications to linguistic theory come directly from Barbara Partee, whose works in this vein also stressed that (a) should at least be the desiderata – the gold standard. (c) is inspired by a variety of work among logicians and philosophers (Quine, Curry and Feys, and others). I present these ideas (informally), and show that taken together, they provide an automatic and striking account of an intriguing empirical puzzle. While one can adopt (a) and (b) and still have a semantics with ‘variables’ (see,e.g., an early account of ‘binding’ of pronouns due to Partee which adopted (a) and could easily have made use of (b)) the adoption of a variable-free semantics yields one further striking prediction in this little corner of data – a prediction which again – is borne out. The larger moral? Not only that these three leading ideas are happy ones for formal semantics to adopt, but that – as semantic theory moves to the future – we should make sure not to lose sight of the enormously rich and subtle body of empirical data that the language gives us to help us hone in on the right tools: small but intriguing corners can tell us quite a bit. (That there is a wealth of surprising, rich and consequential data is also a lesson that Partee has always stressed – for all semanticists are familiar with long lists of incredibly clever and seminal “Partee” sentences, whose moral often pops just from the examples themselves!)
1pm-1:40pm: Florian Schwarz (University of Pennsylvania)
Experimenting with Meaning. Some case-studies in presupposition projection [Slides]
Experimental investigation of meaning-related phenomena can contribute to the enterprise of formal semantics and pragmatics in a number of ways. Beyond empirically testing fine-grained predictions of ever more sophisticated theories, it can also help to shed light on what underlying cognitive representations may correspond to the abstract constructs posited by our theories. Furthermore, it can inform theoretical controversies where one and the same overall predicted interpretive outcome is derived in distinct ways by different approaches, precisely by investigating the factors at play in the process of reaching the final interpretation. Relatedly, experimental work can also contribute to foundational debates about what aspects of a given phenomenon should be explained in terms of domain general cognition, and which need to be encoded linguistically. I illustrate these points through two lines of work on ‘presupposition projection’, i.e., the question of what determines which presuppositions introduced in embedded clauses are inherited by the complex sentence as a whole. First, I briefly review work reported in Schwarz & Tiemann (2017), which finds a pattern of processing costs associated with resolving embedded presuppositions that align with the complexity of structural representations posited by one type of theory, suggesting cognitive correlates of the relevant constructs are at play at some level. Second, I turn to more recent and ongoing work on presupposition projection from conjunction and disjunction, which speak to the core issue of whether projection phenomena can be accounted for in terms of left-to-right processing (Schlenker 2009). While in Mandelkern et al. (2020), we find that conjunction indeed exhibits projection phenomena in line with ‘left-to-right’ unfolding of the linguistic input, but it does so in a rigid way that is at odds with theoretical proposals that merely see this as a default preference due to processing efficiency. In contrast, ongoing work with Alexandros Kalomoiros shows that disjunction entirely lacks any left-to-right asymmetry, which is at odds with existing accounts that posit a uniform impact of deviating from a left-to-right processing preference across connectives. The overall results here thus force us to return to the theoretical drawing board to integrate the explanatory and empirical challenges for a theory of presupposition projection. I close with some more general reflections on the role of experimental approaches in the study of meaning.
1:40pm-2:20pm: Seth Cable (UMass Amherst)
The Remote Past, Recent Past, and Current Past of Cross-Linguistic Formal Semantic Research [Handout]
Almost from its outset, formal (truth-conditional, model theoretic) semantics has striven to capture, elucidate, and explain the variation found between languages in (i) what kinds of meanings can be expressed by certain syntactic constructions, and (ii) how those meanings are compositionally derived. My presentation will begin with a brief historical tour of this line of research, culminating with the landmark publication in 1995 of “Quantification in Natural Languages”, edited by Bach, Jelinek, Kratzer, and Partee. Following this, I will discuss a few more recent examples of interesting and exciting results in this field, such as (potentially) comparatives and other degree constructions, nominal and/or verbal plurality, tense, evidentials, and the expression of ‘de se’ attitudes.
Almost from its outset, formal (truth-conditional, model theoretic) semantics has striven to capture, elucidate, and explain the variation found between languages in (i) what kinds of meanings can be expressed by certain syntactic constructions, and (ii) how those meanings are compositionally derived. My presentation will begin with a brief historical tour of this line of research, culminating with the landmark publication in 1995 of “Quantification in Natural Languages”, edited by Bach, Jelinek, Kratzer, and Partee. Following this, I will discuss a few more recent examples of interesting and exciting results in this field, such as (potentially) comparatives and other degree constructions, nominal and/or verbal plurality, tense, evidentials, and the expression of ‘de se’ attitudes.
2:20-3:00pm: Christopher Potts (Stanford University)
Compositionality or systematicity? [Slides]
The principle of compositionality says roughly that the meaning of a complex syntactic phrase is determined by the meanings of its constituent parts. When linguists seek compositional analyses of linguistic phenomena, what principles guide their investigations, and what higher-level goals are they actually pursuing? In this talk, I’ll begin by trying to answer these questions, and I’ll use those answers to argue that linguists should seriously consider recent models that combine compositional grammars with machine
learning. These models are guided by the principle of compositionality, but they are not strictly speaking
compositional, and compositionality is not a primary objective for their proponents. Rather, the objective of these models is to *generalize* — to make accurate predictions about novel cases. This is a broader goal than compositionality, and adopting it can lead to richer theories of language and language use.
learning. These models are guided by the principle of compositionality, but they are not strictly speaking
compositional, and compositionality is not a primary objective for their proponents. Rather, the objective of these models is to *generalize* — to make accurate predictions about novel cases. This is a broader goal than compositionality, and adopting it can lead to richer theories of language and language use.
3:00pm: Discussion and Closing
Local contact: Mitch Marcus and Charles Yang
Photo of Barbara Partee by Natasha Korotkova