Naive discriminative learning (NDL) is a computational implementation of central ideas of discriminative linguistics, a new theory of language that is under development at the quantitative linguistics lab of the University of Tübingen. Instead of grounding language in a compositional calculus defined over phonemes and morphemes, discriminative linguistics takes inspiration from Shannon's information theory as well as the learning theory of Rescorla and Wagner. Discrimination, not composition, is taken to be fundamental to language and language processing. Discrimination is achieved through error-driven learning, with constant recalibration as experience accumulates over the lifetime. Optimized software implementing discriminative learning is available both for R (ndl package) and for python (pyndl package). This course will comprise lectures in which the principles of discriminative linguistics are introduced. An overview of key results on lexical processing will be presented, ranging from computational modeling of auditory comprehension without mediation by phoneme and word form representations to the modeling of baboon word learning. Lectures will be complemented with lab sessions in which participants receive training in using NDL in the R programming environment.
Languages, like living organisms, evolve and diverge by descent. Substantial insights of what drives these changes has been gained by examining social, geographical and cultural processes, but can cognition also play an important role in the way languages change over time? Here, I will present an overview of a growing body of work exploring how psycholinguistic factors are explaining why words undergo change, and indeed, why they may resist change. Using laboratory based experimental paradigms, psycholinguistic properties of words (such as frequency) can be manipulated and investigated with careful detail and control. Thus, combining this approach with results from cladistics and corpus studies, the aim is to provide robust evidence for cognitive accounts of linguistic change and stability.
The complex nature of human language often makes studying it a challenge. However, linguists and psychologists have utilised the Artificial Language Learning (ALL) paradigm - where miniature novel languages are constructed, learnt and then tested - to explore hypotheses about language acquisition, processing and evolution in a controlled laboratory setting. The aims of this workshop are i) To introduce the ALL paradigm and asses the ways it can be applied to the study of language ii) To present a detailed overview of key findings from ALL studies that investigate grammar, word and orthography learning in adults and children iii) To present ways ALL can be used to study language evolution, alongside computational modelling and corpus based approaches iv) To provide practical instruction on how to design, run and analyse an ALL experiment.
The history of languages can sometimes, but often does not, parallel the genetic history of their speakers. Ever since the publications of seminal papers by Luigi Luca Cavalli-Sforza, Marcus Feldman, and Robert Sokal in the late 1980s and early 1990s, there has been an ongoing discussion about the degree of coevolution between genes and languages on a global scale and the mechanisms responsible for this coevolution. The major question is whether language differences function as barriers to gene flow and thus shape human genetic diversity or whether genetic admixture takes place irrespective of language differences. The apparent coevolution of genes and languages can be merely a consequence of an isolation-by-distance mechanism with decreasing genetic and linguistic exchange over increasing geographical distances. The aim of my talk is to highlight a few of the most interesting cases of congruence (or the lack thereof) between genetic and linguistic relationships. The coevolution between genes and languages is much more complex than previously thought, with demographic and socio-cultural factors at play, and the match between genes and languages is far from universal.
The theme connecting all three of my presentations is the plasticity of the language system in adults and children, and the brain mechanisms that underpin this plasticity. In my first session I will talk about how we learn new words, and the time-course of the processes that occur subsequent to learning, particularly consolidation. I will talk about how sleep makes subtle changes to the way in which our lexical knowledge is represented and discuss theories that predict such changes.
Here, I will focus on the common task in perception of identifying the meaning of an ambiguous word such as “bank”. The lexicon contains information about multiple meanings in cases like this, and there is evidence that these meanings are weighted by frequency. How do we learn to weight these meanings? I will talk about Jenni Rodd’s recent work suggesting that brief experience can change the accessibility of the various meanings, and our collaborative work that suggests that sleep helps to retain these biases. This has surprising consequences for how we think about the act of language comprehension, which may be as much a learning process as an act of perception.
The theories developed in the first two sets of studies can be tested by looking at the learning of quasi-regular systems such as the English past tense, in which some of the information to be learned can be thought of as “regular” and other mappings are treated as exceptions to the rule. For a complementary systems account of language learning, this variation in systematicity is important and makes testable predictions for both learning and consolidation. Here I discuss recent data that tests these predictions and shows the value of memory to the understanding of language, as well as the value of language to the understanding of memory.
The term Grammaticality Judgment and Acceptability Judgment are often used interchangeably although grammaticality and acceptability should not be confused. Grammaticality is a theoretical notion not directly accessible to observation. It is a property of sentences. What we can observe is acceptability, a property of utterances. Though grammaticality is a major factor contributing to acceptability, it is not the only one and it can be overridden by other factors, in particular, but not only, by performance factors. Garden-path sentences as in (1) appear at least momentarily unacceptable despite being grammatical. Grammatical illusions like (2) appear acceptable despite being ungrammatical, in this case because (2) lacks a verb.
(1) The horse raced past the barn fell. (Bever 1970)
(2) The patient the nurse the clinic had hired met Jack. (Frazier 1985)
The talk will address several factors that contribute to acceptability and ways to reduce their impact such that acceptability ratings get as close as possible to grammaticality. Special emphasis will be given to the role of interpretability. Note that utterances are form-meaning pairs. Hence acceptability judgments are inevitably affected by meaning. Grammaticality, on the other hand, concerns the form of sentences and is independent of meaning. A sentence can be completely nonsense and still grammatical. An example is given in (3).
(3) Colorless green ideas sleep furiously. (Chomsky 1957)
We will present an experiment that combines acceptability judgments for which participants were explicitly instructed to ignore meaning and interpretability judgments for which they were instructed to ignore grammaticality. The results show that in principle participants are able to perform these tasks – but not perfectly. We still see an impact of meaning on acceptability and of grammaticality on interpretability judgments.
References:
Bever, T. (1970). The cognitive basis for linguistic structure. In J. R. Hayes (ed.). Cognition and the Development of Language, New York, NY: Wiley and Sons, 279–362.
Chomsky, N. (1957). Syntactic Structures. Mouton, The Hague.
Frazier, L. (1985). Syntactic complexity. In D. R. Dowty, L. Karttunen & A. Zwicky (eds), Natural Language Parsing. Psychological, Computational and Theoretical Perspectives. Cambridge: Cambridge University Press, 129–189.
In this workshop, students will learn to obtain and analyze acceptability judgments by means of formal experiments. We will go through all steps involved in carrying out an experimental study: deriving and formulating a hypothesis, determining an appropriate experimental design, creating experimental stimuli (including fillers and calibration items), setting up the experiment, recruiting participants (as well as dealing with non-cooperative participants), running the experiment, analyzing the data. Since the ultimate goal is to enable participants to carry out their own experiments, the course will break each methodological aspect down into simple steps and provide hands-on practice.
Session 1 will introduce participants to various acceptability judgment tasks (for both absolute and relative acceptability). Session 2 discusses several aspects of experimental designs and procedures. Sessions 3 and 4 focus on the practical aspects of implementing the design and setting up an experiment. Session 5 is devoted to data analysis (with focus on descriptive statistics). We will use R, a programming language and open source software environment for data analysis and data visualization. Finally, Session 6 will address an issue that is particularly prevalent in crowdsourcing: non-cooperative behavior, that is behavior non-compliant with the task. We will present a warning mechanism that efficiently reduces non-cooperative behavior, as well as methods to detect non-cooperative behavior post experiment.
Participants will take out the most of the praxis sessions when bringing their own computers with R and JavaScript already installed (or at least downloaded; if necessary, we will provide help in installing). Code will be provided and explained, no prior technical knowledge is required.
Based on meta-analyses of a large number of hemodynamic studies I will discuss the brain regions in involved in word production. I will also discuss the time course of the component processes of word production and discuss the results of MEG studies investigating the activation time course of brain regions during word production.
I will present the results of meta-analyses of hemodynamic studies on syntactic parsing (comprehension) and encoding (production) and discuss whether comprehension and production share the same neural resources. I will also introduce electrophysiological responses observed for syntactic violations.
I will present data comparing the brain regions involved in sentence-level semantic processing to syntactic processing and discuss the neural underpinnings of different kinds of semantic processing (resolving lexical ambiguities, processing semantic violations, processing metaphors, processing indirect requests and implicatures).
I will present data comparing the brain activations during first (L1) and second (L2) language processing and discuss their implications for the neural representations of the two languages. I will discuss changes in hemodynamic and electrophysiological brain responses in the course of L2 acquisition.
Some Literature:
Hagoort, P. and Indefrey, P. (2014) The neurobiology of language beyond single words. Annual Review of Neuroscience, 37, 347-362.
Indefrey, P. (2006) A Meta-Analysis of Hemodynamic Studies on First and Second Language Processing: Which Suggested Differences Can We Trust and What Do They Mean? Language Learning, 56 (Suppl. 1), 279-304.
Indefrey, P. (2011) The spatial and temporal signatures of word production components: A critical update. Frontiers in Psychology 2:255. doi:10.3389/fpsyg.2011.00255.
Indefrey, P. and Levelt, W.J.M. (2004) The spatial and temporal signatures of word production components. Cognition, 92, 101-144.
Lemhöfer, K. Schriefers, H., and Indefrey, P. (2014) Trust your own errors! The role of subjective syntactic representations in second language comprehension. Journal of Cognitive Neuroscience, 26, 1-17.
Munding, D., Dubarry, A.-S., & Alario, F.-X. (2016). On the Cortical Dynamics of Word Production : A Review of the MEG Evidence. Language, Cognition and Neuroscience, 31, 441-462.
Weber, K. and Indefrey, P. (2009) Syntactic priming in German-English bilinguals during sentence comprehension. Neuroimage 46, 1164-1172.
In a series of four sessions, I will explore the role children may play in language evolution and language transmission. I will start by introducing the methods that have been developed to study these phenomena in the lab, and the cognitive and social biases that are assumed to drive emergence of symbolic communication systems. I will then draw on the developmental literature to explore how this process may be shaped by the different ways in which these biases operate in children and in adults. To this end, I will consider how social and cognitive constraints may affect children’s ability to negotiate meaning during referential communication and to introduce structure during language learning. Finally, taking into account the literature on child-directed speech, I will end with some considerations on why the human ability to teach, defined in its widest sense as demonstration and input enhancement, may be a further factor that should be taken into account when trying to understand the mechanisms of language transmission.
Session 1: Studying Language Evolution in the Lab: Overview and Demonstration
Session 2: Negotiating Meaning – Communicative Constraints in Children and Adults
Session 3: Transmitting Symbolic Signals – Learnability Constraints in Children and Adults
Session 4: Accommodating the Learner – The Role of Teaching in Language Transmission
Day 1: Overview of tone and accent (and relevant little bits of intonation)
A broad general introduction to the range of tonal and accentual phenomena in the world, including discussion of (a) various typological proposals and (b) how European intonation systems fit into the bigger picture.
Day 2: Reconsidering “stress”
What happens when we define ‘stress’ in terms of phonological (‘metrical’ and/or ‘prosodic’) relations between larger constituents, rather than as a phonetic property of individual syllables. Distinguishing ‘metrical’ phenomena (like stress) from ‘autosegmental’ ones (like tone). Another look at the typological and descriptive issues from Day 1.
Languages provide different ways to express real-world situations, their properties, actors and the relations among them. Speakers often have a choice in how to express an idea; for example, they can choose between using the active or passive voice to express the same situation. The choice of one or the other expression reflects how speakers construe (i.e., interpret) different aspects of the world. Even though construal has the potential to provide an insight into the principles underlying conceptualization and language choice, there is a dearth of empirical research investigating construal phenomena. Thus far, most attempts to explain linguistic choices by appealing to alternative construals have relied heavily on analysts' own intuitions about the data. In this talk, I will present a series of experiments that investigate how grammatical encoding interacts with processing, and how individual differences modulate this interaction. By putting the linguistic study of construal into the larger framework of studies on perception, attention and memory, and introducing experimental methodology (including eye-tracking and various individual differences tests), we move towards operationalizing the concept of construal.
According to the central tenets of usage-based linguistics, the categories and constructions of a language emerge from its use. Much recent evidence supports the notion that language users are sensitive to the frequencies of grammatical patterns and the constituents that appear in them. In this talk I will present a series of experiments investigating potential language cues that guide the choice of several grammatical categories: case and aspect in Polish and articles and tense in English. In these studies, we use (among other techniques) keystroke logging to track how participants change texts to make them sound more acceptable. Keylogging provides a new, online approach to observe writing processes and the keylogging data offer a detailed insight into participants’ linguistic decision-making processes. For example, certain types of changes that are more frequent across participants are reflected in faster typing speed. In addition, we investigate how linguistic performance is related to individual differences.
Language is one of the central topics in the investigation of human cognition. Thus unsurprisingly, research into this highly complex phenomenon has been an integral part of not only linguistics, but also disciplines like psychology and computer science. In this talk, I will discuss how an interdisciplinary approach combining insights from these distinct disciplines can inform our understanding of basic principles underpinning language learning, use, and representation.
The workshop will review methods that can be used to examine how 1- to 3-year-olds comprehend various language structures. It will briefly describe offline methods such as picture-pointing or enactment tasks, but the focus will be on the real-time (online) methods of preferential looking and looking-while-listening (visual world). Various practical aspects of using these paradigms will be discussed, including the technical setup of the apparatus. The workshop will provide an introduction the R package eyetrackingR, wich can be used to evaluate the preferential looking data.
The ‘classical’ approach to morphology postulates that knowing and generalising relations between words depends, essentially, on knowing rules: categorical operations which create structured representations (walk, Xed → [walk][ed]). Many so-called dual-mechanism theories additionally distinguish between structured rule-based forms (e.g., for regular, productive operations) and whole-form storage (e.g., for irregulars). Alternatively, within analogical, connectionist, and stochastic approaches, it has been proposed that a single mechanism, which is inherently graded and sensitive to frequency and similarity can generalise and process complex forms without explicit morphological structure.
In these sessions, we will review a number of experimental and computational studies in word recognition and morphological processing, which have attempted to adjudicate between these broad theoretical positions. Special emphasis will be given to (a) different computational frameworks of word recognition and morphological knowledge, with demonstrations of running models (e.g., interactive activation, Albright’s MGL, Skousen’s AML, connectionism); (b) linguistic phenomena and distinctions that are ‘essentially morphological’, such as conjugation classes or the distinction between inflection and derivation; and (c) cross-linguistic studies, particularly those conducted in Romance and Semitic languages.