Carnegie Mellon University

NASSLLI 2018 @ CMU - June 23-29

North American Summer School on Logic, Language, and Information

NASSLLI 2018 logo

NASSLLI Student Session

Student poster sessions will take place in the coffee break area (Baker Hall A50) during the lunch block (11:45-1:30), Tuesday through Thursday


Tuesday, June 26

Code-mixing occurs when people use multiple languages to communicate. We want to understand code-mixing in a wider context of a dialogue between two people, and effectively learn about the bilingual mind and the linguistic and social triggers of code-mixing. We have developed a Spanish-English goal-oriented dialogue system where we manipulate the styles of code-mixing in the generation on three levels: syntactic, lexical, and social. We present results from releasing this system on a crowdsourcing platform, in order to gather conversations of our chatbot with crowd-sourced bilinguals. By examining the humans’ responses, we can learn if certain styles of mixing are more effective than others.
No description.

I argue that a correct formulation of inexact perceptual knowledge must take into account the dynamics of inexact observations. I develop a novel version of Dynamic Epistemic Logic (DEL) which allows for inexact epistemic updates, and discuss its connection with hybrid logic. My account aims to dissolve the philosophical problems related to inexact knowledge and to provide a more realistic logic of perceptual knowledge.

Most models of decision making do not consider the number of hypotheses (or other types of choices) to be a relevant factor in hypothesis choice. In reality, the number of hypotheses considered to be viable options may increase or decrease due to new evidence or a change in circumstances. This poster presents a Bayesian network model that represents the belief states of agents within a social community that demonstrates how incoming information from epistemic neighbors, in conjunction with evidence taken in from each agent’s own experiences in the world, contributes to the agents choosing what they each consider to be the best hypothesis from the number of hypotheses they believe to be viable options.
Chinese relative clauses may appear in either pre-Dem or post-Dem position in the nominal structure (i.e. <RC>—Dem—Num—CL—<RC>—N). The general consensus is that the two structural positions correlate with the restrictive/non-restrictive nature of the relative. This poster argues against this empirical generalisation and proposes that the relative/non-relative distinction correlates with the different uses of Dem (deictic or non-deictic), depending on whether or not the indexical position of Dem (as an anaphoric definite, à la Schwarz 2009, Jenks 2018) is occupied by a demonstration gesture. The proposal explains why established diagnostics for the restrictive/non-restrictive distinction, when applied to Chinese, yield contradictory results, and furthermore accounts for an unnoticed asymmetry between proximal Dem (zhe ‘this’) and the non-proximal Dem (na ‘that’) in relative formation.
The particles 'also' and 'too' are the standard way of signaling additivity in assertions and polar questions. It has been suggested that they can't appear in wh-questions without triggering a so-called showmaster interpretation, though. I present new data challenging this generalization and propose a way of generalizing additivity presuppositions that can account for the resulting, more differentiated empirical picture.

Wednesday, June 27

Supposition and conditionals appear closely related. For example, (1) and (2) seem to play similar roles in discourse:
  1. Suppose that the butler did it. Then the gardener is innocent.
  2. If the butler did it, then the gardener is innocent.
Suppositional Theories appeal to supposition in providing an account of the natural language conditional (e.g., Quine (1952); Mackie (1973); Adams (1975); Edgington (1995), a.o.). For example, here is J.L. Mackie:

“The basic concept required for the interpretation of if-sentences is that of supposing […] To assert ‘If p, q’ is to assert q within the scope of the supposition that p” (1973, 92-93).

This project develops a suppositional theory of conditionals. However, it differs from extant theories in (i) arguing for a precise semantic connection between instructions to suppose and conditional antecedents, and (ii) providing novel linguistic data in favor of that theory.
Distributive numerals in Mandarin Chinese (NumNum) are formed by reduplicating a numeral + classifier combination. NumNum modifies a plural event, specifying the cardinality of a participant of each subevent. Adopting a syntactic analysis of NumNum as taking parasitic scope, we present our semantics with two key ingredients: (1) a framework with both sets and sums of events, and (2) an underspecified thematic relation 'Participant' in the metalanguage, as in Cable (2014). We defend our analysis with evidence from sentences with two NumNums and NumNum's interaction with reciprocals.
I use scalar implicatures as a lens through which to examine two compositional distributional models: the Multiplication Model (Mitchell and Lapata, 2010) and CCDS (Coecke et al., 2010). I will look at two opposing views regarding the derivation of scalar implicatures, and will illustrate how they would respectively be integrated within these distributional models.
Computational linguists have long studied various logic forms for capturing essential semantic information carried by narratives. Among these, discourse representation structure (DRS) form is designed to acquire the entities, entities’ property, events, event types, the occurring time of events, and event arguments. In this poster​, we describe a system called Text2DRS ​​that takes English narrative as an input and outputs a DRS in Neo-Davidsonian style. In this regard, it is similar to ​earlier system ​Boxer​ by Johan Bos​,​​ which is an open-domain NLP tool for semantic analysis of a text and produces a DRS for a given narrative. Text2DRS provides additional information in comparison to Boxer. For example, Text2DRS relies on lexical resource VerbNet for annotating the specific relations between relevant entities and events mentioned in the narrative using the verb classes and thematic roles of VerbNet.

I consider how to derive free relatives – e.g. John eats [DP what Mary eats] – in Minimalist Grammars (Stabler 1997).  Free relatives are string-identical to indirect questions – e.g. John wonders [CP what Mary eats].  An analysis of free relatives as nominalised indirect questions is easy to implement, but empirical evidence points instead to wh-words ‘reprojecting’ in free relatives (Donati 2006).  I propose two innovations in Minimalist Grammars to implement a reprojection analysis: (i) feature recycling and (ii) a Reproject operation.

Thursday, June 28

Kit Fine (2017) distinguishes between exact truthmaking, which requires that every part of the truthmaker be relevant to making the sentence true, and inexact truthmaking, which lacks this requirement (and is akin to situation semanticists’ notion of (persistent) verification). He argues that because we can define the latter in terms of the former, but not vice versa, we should take exact truthmaking to be the fundamental notion. But I argue this this gets things backwards: inexact truthmaking can be used to define exact truthmaking, but not vice versa. Truthmaker semantics should be built on a foundation of inexact truthmaking.
This study aims to analyze the use of 30 common English acronyms in Chinese text/speech, using several newspaper corpora and Google books to represent text and TV/radio corpora to represent speech. We first use the percentage of English acronyms (prctEn) to measure the degree of acronym usage. prctEn is defined as the frequency of English acronyms over the summed frequency of its English and Chinese terms: prctEn (GDP) = count(GDP) / count(GDP+国内生产总值, guomin shengchan zongzhi, literal translation of "Gross Domestic Product"). The result first shows an increasing trend for prctEn diachronically. Several factors (frequency of the concept, length of Chinese translation, etc.) are predicted to <chrome_find class="find_in_page findysel">influence prctEn of a given acronym in the linear mixed-effects model. The length of the Chinese translation is the single most important reason for which speakers would prefer the English acronym over its Chinese translation, i.e. the shorter the better.
In modal logic, modal expressions can be classified according to their quantificational force (i.e. must and may). Also, among necessity modals, there exists a strength distinction (e.g. must and should). This poster entertains a possibility laid down by these two parameters. In particular, in Chinese, the dual counterpart of should and a logically stronger version of may are realized as you-liyou 'have-reason’, the semantics of which can be naturally captured under Kratzerian approach to modality.
No description.

Constraint answer set programming (CASP) is a paradigm designed to tackle several issues of answer set programming tools. Recently, a method is proposed to find solutions to CASP programs by using satisfiability modulo theories (SMT) solvers.

This SMT-based approach is implemented in a system called EZSMT, which often outperforms its peers. Yet, it has several limitations. For instance, it is unable to process a large class of programs called "non-tight". This restriction does not allow users to express, for example, reachability relations between cities connected by roads on a map. Solving non-tight programs is important, because they are crucial in many applications.

Here we extend EZSMT for non-tight programs and constraints over mixed domains of reals and integers. We also extend EZSMT to allow the computation of multiple solutions. Our experimental analysis shows that the extended EZSMT system is a compatible tool compared to state-of-the-art CASP solvers CLINGCON and EZCSP.

No description.

Presenting Sponsor:

Citi Ventures