Semantics and Linguistic Theory (SALT) 20 April 29 - May 1, 2010 University of British Columbia and Simon Fraser University Vancouver, BC, Canada Collected Abstracts Invited Speakers A possible worlds semantics for (illocutionary) evidentials Martina Faller The precise relationship between evidentiality, the linguistic marking of the speaker's source of information for a claim, and epistemic modality, the marking of the speaker's evaluation of the truth of a proposition in terms of necessity, possibility or degree of certainty, is not yet fully understood. While these two phenomena are clearly distinct conceptually, it is not always possible to establish two distinct categories empirically in a given language because the two concepts are often expressed simultaneously by a single element. For example, English epistemic must expresses that the speaker s considers the proposition expressed p a necessity and in addition indicates that s has inferred p and does not have direct evidence for it. The two concepts are furthermore closely related in that the type of source information a speaker has will to a great extent determine the speaker's evaluation of the truth of the proposition. Within formal semantics, evidentials are often analyzed as quantifiers over possible worlds with evidential presuppositions, that is, as a kind of epistemic modal. In this talk I will explore to what extent such an analysis is viable for evidentials that do not contribute to the main proposition expressed/at-issue content such as the evidentials of Cuzco Quechua (CQ). CQ has four evidentials which I have ananlyzed in previous work as contributing to the illocutionary level of meaning. The Direct indicates that the speaker s has direct evidence for p, the Reportative that s was told that p, the Conjectural that s conjectures that p, and the perceived evidence Inferential that s infers that p perceived evidence. I will argue that all four can be analyzed as quantifiers over possible worlds, because even the non-inferential evidentials involve some degree of inference. However, since their main contribution is to indicate what type of evidence this inference is based on, I will argue that the evidential requirement should not be analyzed as a presupposition. Instead, I propose that evidentials assert the existence of a set of facts acquired by direct means or from reports, from which p follows. If we assume that, in general, speakers come to believe a proposition p on the basis of some evidential event via a (possibly minimal) amount of reasoning, the difference between epistemic modals and evidentials lies in what part of this process they focus on. Epistemic modals focus on the inferential part, and evidentials on the evidential event. Unmodified assertions make no reference to this process at all but simply assert the proposition itself. The acquisition of meaning: Evidentiality in semantics and cognition Anna Papafragou How do semantics and cognition make contact during language learning? This talk addresses this question by investigating the acquisition of evidentiality (the linguistic encoding of information source) and its relation to children’s evidential reasoning. I present data from a series of experimental studies with children learning Turkish and Korean (two languages with evidential morphology) and English (a language without grammaticalized evidentiality) in order to test two hypotheses: (a) the acquisition of evidentiality is complicated by the subtleness and abstractness of the underlying concepts; (b) learning a language which systematically (e.g., grammatically) marks evidential categories might affect early reasoning about sources of information. The experiments show that the production and comprehension of grammaticalized evidentials can pose considerable difficulty to young learners; nevertheless, these problems are not (necessarily) conceptual in nature since the same learners successfully reason about sources of information in non-linguistic tasks. Furthermore, children’s ability to reason about sources of information proceeds along similar lines in diverse language-learning populations and is not tied to the acquisition of the linguistic markers of evidentiality in the exposure language. I discuss implications of these findings for the relationship between linguistic-semantic and conceptual representations during development. Emergent expressivity, Christopher Potts (Stanford) SALT 20, UBC, April 29 – May 1, 2010 Meaning and use Kaplan (1999) writes, “When I think about my own understanding of the words and phrases of my native language, I find that in some cases I am inclined to say that I know what they mean, and in other cases it seems more natural to say that I know how to use them.” At the ‘meaning’ end of the spectrum, we might be tempted to place common nouns and boolean connectives; at the ‘use’ end, expressives and (other) indexicals. The overarching question for this talk is, How would our investigations proceed if we had information only about use conditions, with no direct access to propositional content? What would our theories look like? What inferences would we make about the underlying meanings? In the spirit of these questions, I’ve structured this abstract (and will structure the talk) around two mystery elements, M1 and M2, for which I supply a large amount of information about use conditions, in the form of quantitative evidence drawn from large corpora. I’ve also identified some elements with similar distributional properties — similar use conditions — to guide us. Data This abstract involves data from Twitter (10 million ‘tweets’, each at most 140 characters; 10.2 million words) and from a collection of online product reviews (560,000 reviews; 39.7 million words; Davis and Potts 2009). Twitter has evolved into a tool for broadcasting information, so the contexts tend to be purely information-sharing. The reviews mix information-sharing with argumentation about the relative merits of the products under review. The talk will supplement these corpora with data drawn from OpposingViews.com (argumentative), Goodreads.com (reviews and social networking), and Thomas et al.’s (2006) Congressional Speech Data corpus (argumentative). Mystery item M1 Figures 1–2 depict the distribution of a range of lexical items in the corpora, with M1 in the rightmost panels. In each case, M1 appears to be a milder version of the negative scalar modifiers, and it stands opposed to the positive scalar modifiers. The figures’ captions provide additional details about the plots. The high-level picture is that M1, like bad and terrible, is used more heavily in contexts of negative sentiment, conflict, and disagreement. What is the source of these usage conditions? Mystery item M2 Figures 3(a)–3(b) show that negative expressives amplify already negative messages. Intensional domain wideners (Rawlins 2008) like who on earth are negatively biased, and their expressive variants increase this bias. Similarly, the intensive totally has an expressive counterpart fucking, which veers more sharply to the negative end of the spectrum. Figure 3(c) depicts M1 occurring with M2 elements in its scope. Like the lexical negative expressives, M2 amplifies the negative bias seen with M1 alone. The evidence also suggests that repeating M2 items increases the negative effect (Potts 2007). What is at the root of this pragmatic connection between expressives and M2 items? well excellent(ly) terribl(e|y) bad(ly) M1 1.0 1.0 1.0 1.0 1.0 0.82 0.75 0.8 0.8 0.8 0.8 0.8 0.67 0.65 0.62 0.6 0.6 0.6 0.6 0.6 0.38 0.33 0.35 0.4 0.4 0.4 0.4 0.4 0.25 0.18 0.2 0.2 0.2 0.2 0.2 0.0 0.0 0.0 0.0 0.0 :−( :−) :−( :−) :−( :−) :−( :−) :−( :−) Probability given emoticon Probability given Clausemate emoticon Figure 1: Twitter (information-sharing contexts): Emoticons are frequent on Twitter, where they are used as general markers of sentiment. These panels restrict attention to uses of the items in question where they have clausemate positive (smiley) or negative (frownie) emoticons. Positive language correlates with positive emoticons, negative language with negative emoticons, and the degree of biased aligns well with intuitions about lexical strength. M1 emerges as mildly negative by this metric. 1 Emergent expressivity, Christopher Potts (Stanford) SALT 20, UBC, April 29 – May 1, 2010 well excellent(ly) terribl(e|y) bad(ly) M1 ! ER = 0.21 ER = 1 ER = −1 ER = −0.67 ER = −0.29 ! 0.4 0.4 0.4 0.4 0.4 ! ! 0.3 0.3 0.3 0.3 0.3 ! ! ! ! ! ! ! ! ! 0.2 ! 0.2 0.2 0.2 0.2 ! ! ! ! Probability ! ! 0.1 0.1 ! 0.1 ! 0.1 ! 0.1 ! ! −2 −1 0 1 2 −2 −1 0 1 2 −2 −1 0 1 2 −2 −1 0 1 2 −2 −1 0 1 2 Rating (centered around 0) Figure 2: Review corpora (information-sharing and argumentative contexts): These panels depict relative frequencies across the five-star rating scale R, which is centered around 0 to reflect the fact that 1-2 star reviews are negative, 4-5 star reviews are positive, and 3-star reviews are lukewarm (Potts and Schwarz 2010). The expected rating (ER) values are weighted averages: r R rPr(r). The higher (lower) the ER value, the more positive (negative) the item. M1 has a negative bias, reliable∈ but mild compared with terrible. It is comparable to well on the positive side. wh− (earth|world) totally ADJ totally ADJ M1 scoping over a single M2 1.0 0.5 0.5 ! ER = −0.74 ER = −0.1 ER = −0.51 0.4 0.8 0.4 0.4 ! 0.6 0.3 0.52 0.3 0.3 0.48 ! ! ! ! ! 0.2 0.2 ! ! 0.4 ! 0.2 ! ! ! 0.1 0.1 ! 0.2 ! 0.1 0.0 0.0 0.0 −2 −1 0 1 2 −2 −1 0 1 2 :−( :−) −2 −1 0 1 2 wh− (hell|fuck) fuckin(g) ADJ fuckin(g) ADJ M1 scoping over multiple M2 1.0 0.5 Probability ! Probability Probability ! 0.5 ER = −0.87 ER = −0.85 ER = −0.53 0.4 0.8 0.4 0.4 0.63 ! Probability given emoticon Probability given 0.3 0.6 0.3 0.3 ! ! 0.37 0.2 0.2 ! 0.4 ! 0.2 ! ! ! ! ! 0.1 0.1 ! 0.2 ! 0.1 ! 0.0 0.0 0.0 −2 −1 0 1 2 −2 −1 0 1 2 :−( :−) −2 −1 0 1 2 Rating (centered around 0) Rating (centered around 0) Clausemate emoticon Rating (centered around 0) (a) The negative bias of (b) Whereas intensive totally has only a slight negative (c) M1 and M2 as clause- normal intensional domain bias, fucking is used almost exclusively in negative con- mates.