The 60th Language Lunch
Date: 2017-12-07
Location: G.07 Informatics Forum
The evolution of monotonicity in the semantics of gradable adjectives
Fausto Carcassi; s1569804@sms.ed.ac.uk
Gradable adjectives like tall can be used in two ways: measure (e.g. John is 180cm tall) and bare (e.g. John is tall). In bare uses the adjective says that the subject falls within a certain interval in the adjective’s scale, which we call the adjective’s bare interval. We show that bare intervals are monotonic according to the standard semantics (Kennedy and McNally 2005), i.e. there is a degree d such that the bare interval includes exactly the degrees greater or the degrees lesser than d.rnrnWe present three computational models to explain the evolution of monotonicity. The first model uses the Iterated Learning paradigm (Kirby, Griffiths, and Smith 2014) to study how bare intervals evolve under a pressure for learnability alone in a population of literal agents. We show that while bare intervals become monotonic, they also degenerate, i.e. end up covering either nothing or the whole degree’s scale. This is unlike bare intervals of natural languages. The second model studies the effects of adding a pressure for communicative accuracy. Bare intervals stop being degenerate and become non-monotonic, which is still different from natural language adjectives. Finally, in the third model we implement more sophisticated, pragmatic agents using the Rational Speech Act modelling paradigm (Goodman and Frank 2016). This allows agents to calculate scalar implicatures. Languages in the third model finally evolve non-degenerate monotonic meanings resembling bare intervals of real languages.
Can L2 Speakers Acquire New Morphological Distinctions? Evidence from temporal Morphological Production in Mandarin Speakers of English
Qingyuan Gardner; s1360390@sms.ed.ac.uk
L2 speakers of English from different L1 backgrounds have been widely observed to exhibit inconsistent production of L2 morphological inflections (Lardiere, 1998, 2000; Hawkins & Liszka, 2003). The current set of experiments focuses on L1 Mandarin speakers of English, whose native language is tense-free and does not use a morphological system to indicate temporal properties of events. Through 2 spoken production studies and 1 written production study, we investigated the mechanisms which underlie such inconsistent production from a psycholinguistic perspective. The experiments comprised of online picture description tasks which required participants to respond under time constraints, and offline language proficiency assessments. Our findings reject the hypotheses that L2 speakers do not represent L2 specific linguistic features and that inconsistent morphological production is solely an articulatory problem (Hawkins & Chan, 1997; Goad, White & Steele, 2003). The findings also indicate problematic morphological encoding in L2 speakers.
The development of English tense and agreement morphology in Welsh-English bilingual children with and without Specific Language Impairment
Hyowon Kwon; None; None
This study investigated whether third person singular (3SG -s) and past tense accuracy and error types reveal distinctive developmental patterns of agreement and tense acquisition in different age and language impairment groups of Welsh L1 children with English as a second language (L2). A group of Welsh-English bilingual typically developing (L2-TD) older (6;7-9;0-year-old) and younger (4;6-6;4-year-old) children and a group of young (4;6-6;2-year-rnold) children with SLI age-matched to the younger L2-TD group were administered the screening component of the Test of Early Grammatical Impairment (Rice & Wexler, 2001). The results indicated that the three groups differed in their production of 3SG -s and regular past tense. However, there were no differences in terms of accuracy with the irregular past tense verbs. The L2-SLI children produced similar error types to the younger L2-TD children, who differed from their older L2-TD peers. At the same time, vocabulary size and morphophonology contributed to children’s performance. We discuss these results within current accounts of language development and impairment.
From Characters to Words to in Between: Do We Capture Morphology?
Clara,Vania; c.vania@ed.ac.uk
Naomi Saphra; s1477768@sms.ed.ac.uk
Words can be represented by composing the representations of subword units such as word segments, characters, and/or character n-grams. While such representations are effective and may capture the morphological regularities of words, they have not been systematically compared, and it is not understood how they interact with different morphological typologies. On a language modeling task, we present experiments that systematically vary (1) the basic unit of representation, (2) the composition of these representations, and (3) the morphological typology of the language modeled. Our results extend previous findings that character representations are effective across typologies, and we find that a previously unstudied combination of character trigram representations composed with bi-LSTMs outperforms most others. But we also find room for improvement: none of the character-level models match the predictive accuracy of a model with access to true morphological analyses, even when learned from an order of magnitude more data.
