Linguistics Commons^™

Generative Linguistics And Neural Networks At 60: Foundation, Friction, And Fusion, Joe Pater

Joe Pater

Text-Speech Alignment: A Robin Hood Approach For Endangered Languages, Claire Bowern, Rikker Dockum, Sarah Babinski, Hunter Craft, Anelisa Fergus, Dolly Goldenberg

Yale Day of Data

Forced alignment automatically aligns audio recordings of spoken language with transcripts at the level of individual sounds, greatly reducing the time required to prepare data for linguistic analysis. However, existing algorithms are mostly trained on a few well-documented languages. We test the performance of three algorithms against manually aligned data on data from a highly endangered language. At least some tasks, unsupervised alignment (either based on English or trained from a small corpus) is sufficiently reliable for it to be used on legacy data for low-resource languages. Descriptive phonetic work on vowel inventories and prosody can be accurately captured by ...

Structural Justice: A Critical Feminist Framework Exploring The Intersection Between Justice, Equity And Structural Reconciliation., Camille Burnett, Michael Swanberg, Ashley Hudson, Donna Schminkey

Journal of Health Disparities Research and Practice

Violence against women is a human rights violation (UN, 2006). It affects the health of women globally (UN, 2009) and its elimination is at the heart of many international and national goals. Intimate partner violence (IPV), one of the most common forms of gender-based violence, affects one in three women worldwide (WHO, 2013). The consequences of IPV create negative health outcomes for women that diminish their quality of life and their overall well-being. Abused women access community supports such as shelters to seek safe refuge from the abuse and restore their lives. While shelters play an extensive role in helping ...

Identifying Participation Of Individual Verbs Or Verbnet Classes In The Causative Alternation, Esther Seyffarth

Proceedings of the Society for Computation in Linguistics

Verbs that participate in diathesis alternations have different semantics in their different syntactic environments, which need to be distinguished in order to process these verbs and their contexts correctly. We design and implement 8 approaches to the automatic identification of the causative alternation in English (3 based on VerbNet classes, 5 based on individual verbs). For verbs in this alternation, the semantic roles that contribute to the meaning of the verb can be associated with different syntactic slots. Our most successful approaches use distributional vectors and achieve an F1 score of up to 79% on a balanced test set. We ...

Can Entropy Explain Successor Surprisal Effects In Reading?, Marten Van Schijndel, Tal Linzen

Proceedings of the Society for Computation in Linguistics

Human reading behavior is sensitive to surprisal: more predictable words tend to be read faster. Unexpectedly, this applies not only to the surprisal of the word that is currently being read, but also to the surprisal of upcoming (successor) words that have not been fixated yet. This finding has been interpreted as evidence that readers can extract lexical information parafoveally. Calling this interpretation into question, Angele et al. (2015) showed that successor effects appear even in contexts in which those successor words are not yet visible. They hypothesized that successor surprisal predicts reading time because it approximates the reader’s ...

Modeling Clausal Complementation For A Grammar Engineering Resource, Olga Zamaraeva, Kristen Howell, Emily M. Bender

Proceedings of the Society for Computation in Linguistics

We present a grammar engineering library for modeling objectival declarative clausal complementation patterns attested cross-linguistically. Our primary contribution is positing a set of syntactico-semantic analyses couched within a variant of the HPSG syntactic formalism and integrating them with a variety of phenomena already implemented in a grammar engineering toolkit. We evaluate the addition to the system on testsuites from genetically diverse languages that were not considered during development.

Do Rnns Learn Human-Like Abstract Word Order Preferences?, Richard Futrell, Roger P. Levy

Proceedings of the Society for Computation in Linguistics

RNN language models have achieved state-of-the-art results on various tasks, but what exactly they are representing about syntax is as yet unclear. Here we investigate whether RNN language models learn humanlike word order preferences in syntactic alternations. We collect language model surprisal scores for controlled sentence stimuli exhibiting major syntactic alternations in English: heavy NP shift, particle shift, the dative alternation, and the genitive alternation. We show that RNN language models reproduce human preferences in these alternations based on NP length, animacy, and definiteness. We collect human acceptability ratings for our stimuli, in the first acceptability judgment experiment directly manipulating ...

Quantifying The Relationship Between Child And Caregiver Speech Using Generalized Estimating Equations: The Case Of 'Only', Lindsay Hracs

Proceedings of the Society for Computation in Linguistics

One of the difficulties involved in modelling longitudinal data is that repeated measurements over time introduce a violation of independence. Standard Generalized Linear Models are not robust to this violation. However, Generalized Estimating Equations (GEEs) take correlations between data points into consideration making them useful for such tasks. This paper examines the use of GEEs to model the relationship between child-directed and child-produced speech, focusing on the role of input in the acquisition of only. The study shows that the frequency of occurrence of only in child-directed speech is a significant predictor of the frequency of occurrence of only in ...

Guess Who’S Coming (And Who’S Going): Bringing Perspective To The Rational Speech Acts Framework, Carolyn Jane Anderson, Brian W. Dillon

Proceedings of the Society for Computation in Linguistics

We present a Rational Speech Acts approach to modeling how conversation participants reason about perspectival expressions. The interpretation of perspectival expressions, such as the motion verbs 'come' and 'go', depends on the point-of-view from which they are evaluated. In order to interpret a perspectival expression, the listener must jointly reason about the speaker’s intended message and their choice of perspective. We propose a Bayesian approach to this inference problem and describe an extension of the Rational Speech Acts model that incorporates perspective. We lay out three sets of predictions that this model makes relating to the lexical semantics of ...

The Organization Of Sound Inventories: A Study On Obstruent Gaps, Sheng-Fu Wang

Proceedings of the Society for Computation in Linguistics

This study explores the organizing principles of sound inventories by examining attested one-segment gaps in obstruent inventories. Models based on different theories of inventory organization are built and compared in a computational task where models make a binary decision to identify gaps and attested sounds. Results show that segment markedness, defined either in terms of grounded phonetic properties or typological frequencies, is a good predictor of whether a segment is likely to be gapped in an inventory. On the other hand, whether an attested segment, compared to a gapped segment, makes the feature representation more symmetric or economical, is not ...

Local Processes Of Homophone Acquisition, Deniz Beser, Spencer Caplan

Proceedings of the Society for Computation in Linguistics

The Naive Generalization Model (NGM) (Caplan, 2018) explains word learning phenomena as grounded in the local, dynamical process of category formation. A range of experimental evidence (Xu and Tenenbaum, 2007; Spencer et al., 2011; Lewis and Frank, 2018) supports the NGM over prior models of word learning such as Bayesian inference (Xu and Tenenbaum, 2007). Despite such progress, a number of theoretical phenomena remain unaddressed by previous accounts. In this paper, we present a novel extension to NGM which offers a strong fit to and explanation of experimental data on homophone acquisition (Dautriche et al., 2016).

Re(Current) Reduplication: Interpretable Neural Network Models Of Morphological Copying, Colin Wilson

Proceedings of the Society for Computation in Linguistics

No abstract provided.

Targeted Syntactic Evaluation Of Language Models, Rebecca Marvin, Tal Linzen

Proceedings of the Society for Computation in Linguistics

We present a dataset for evaluating the grammatical sophistication of language models (LMs). We construct a large number of minimal pairs illustrating constraints on subject-verb agreement, reflexive anaphora and negative polarity items, in several English constructions; we expect LMs to assign a higher probability to the grammatical member of each minimal pair. An LSTM LM performed poorly in many cases. Multi-task training with a syntactic objective improved the LSTM’s accuracy, which nevertheless remained far lower than the accuracy of human participants. This suggests that there is considerable room for improvement over LSTMs in capturing syntax in an LM.

A Conceptual Spaces Model Of Socially Motivated Language Change, Heather Burnett, Olivier Bonami

Proceedings of the Society for Computation in Linguistics

This paper outlines a formal model of socially motivated language change which unites insights from identity-oriented theories of language change with formal theories of language use and understanding. We use Gärdenfors's (2000) Conceptual Spaces framework to formalize socially motivated ideological change and use signaling games with an iterated best response solution concept (Franke, 2009; Frank and Goodman, 2012) to formalize the link between ideology, linguistic meaning and language use. We then show how this new framework can be used to shed light on the mechanisms underlying socially-motivated change in French grammatical gender.

Learnability And Overgeneration In Computational Syntax, Yiding Hao

Proceedings of the Society for Computation in Linguistics

This paper addresses the hypothesis that unnatural patterns generated by grammar formalisms can be eliminated on the grounds that they are unlearnable. I consider three examples of formal languages thought to represent dependencies unattested in natural language syntax, and show that all three can be learned by grammar induction algorithms following the Distributional Learning paradigm of Clark and Eyraud (2007). While learnable language classes are restrictive by necessity (Gold, 1967), these facts suggest that learnability alone may be insufficient for addressing concerns of overgeneration in syntax.

An Incremental Iterated Response Model Of Pragmatics, Reuben Cohn-Gordon, Noah Goodman, Christopher Potts

Proceedings of the Society for Computation in Linguistics

Recent Iterated Response (IR) models of pragmatics conceptualize language use as a recursive process in which agents reason about each other to increase communicative efficiency. These models are generally defined over complete utterances. However, there is substantial evidence that pragmatic reasoning takes place incrementally during production and comprehension. We address this with an incremental IR model. We compare the incremental and global versions using computational simulations, and we assess the incremental model against existing experimental data and in the TUNA corpus for referring expression generation, showing that the model can capture phenomena out of reach of global versions.

Simultaneous Learning Of Vowel Harmony And Segmentation, Ezer Rasin, Nur Lan, Roni Katzir

Proceedings of the Society for Computation in Linguistics

No abstract provided.

Linguistic Alignment Is Affected More By Lexical Surprisal Rather Than Social Power, Yang Xu, Jeremy Cole, David Reitter

Proceedings of the Society for Computation in Linguistics

No abstract provided.

Quantifying Structural And Lexical Constraints In Pp Ordering Typology, Zoey Liu

Proceedings of the Society for Computation in Linguistics

No abstract provided.

Q-Theory Representations Are Logically Equivalent To Autosegmental Representations, Nick Danis, Adam Jardine

Proceedings of the Society for Computation in Linguistics

We use model theory and logical interpretations to systematically compare two competing representational theories in phonology, Q-Theory (Shih and Inkelas, 2014, forthcoming) and Autosegmental Phonology (Goldsmith, 1976). We find that, under reasonable assumptions for capturing tone patterns, Q-Theory Representations are equivalent to Autosegmental Representations, in that any constraint that can be written in one theory can be written in another. This contradicts the assertions of Shih and Inkelas, who claim that Q-Theory Representations are different from, and superior to, Autosegmental Representations.