Interactive task learning studies situations in which a teacher (task instructor) interacts with a learner (task executor) to perform a novel task in an embodied environment. To successfully interpret the teacher's utterances, the learner has to perform interactive symbol grounding: it must update its prior beliefs about the mapping from symbols to visual referents each time the teacher speaks. Interactive symbol grounding is even more challenging if the learner starts out unaware of concepts that are critical to task success. In that case, the learner must use the embodied conversation to discover and adapt to unforeseen possibilities, and so must cope with a continuously expanding hypothesis space and hence a non-stationary domain model, requiring structure-level updates during interaction. In this paper, we propose a neuro-symbolic model for learning dialogue strategies for achieving interactive symbol grounding. In particular, we study the effects of enriching the model with symbolic reasoning that captures the valid consequences of quantifiers (e.g., both, every). Our hypothesis is that utilizing such reasoning makes interactive task learning more data efficient. We test this empirically via a task of interactive reference resolution, in which the learner must jointly learn a grounding model and a policy for querying the teacher to enhance its accuracy in grounding. Our results show that a learner that exploits such symbolic reasoning for both decision-making and grounding is more data-efficient than learners that ignore such linguistic insights.
A recent research direction in computational linguistics involves efforts to make the field, which used to focus primarily on English, more multilingual and inclusive. However, resource creation often remains a bottleneck for many languages, in particular at the semantic level. In this article, we consider the case of frame-semantic annotation. We investigate how to perform frame selection for annotation in a target language by taking advantage of existing annotations in different, supplementary languages, with the goal of reducing the required annotation effort in the target language. We measure success by training and testing frame identification models for the target language. We base our selection methods on measuring frame transferability in the supplementary language, where we estimate which frames will transfer poorly, and therefore should receive more annotation, in the target language. We apply our approach to English, German, and French – three languages which have annotations that are similar in size as well as frames with overlapping lexicographic definitions. We find that transferability is indeed a useful indicator and supports a setup where a limited amount of target language data is sufficient to train frame identification systems.
Vaidya et al. (2019) discuss argument alternations in Hindi complex predicates, and propose an analysis within an LTAG framework, comparing this with an LFG analysis of complex predicates. In this paper I clarify the inadequacies in existing LFG analyses of complex predicates, and show how the LFG+glue approach proposed by Lowe (2015) can both address these inadequacies and provide a relatively simple treatment of the phenomena discussed by Vaidya et al. (2019).
No matter how comprehensively corpus builders design their annotation schemes, users frequently find that information is missing that they need for their research. In this methodological paper I describe and illustrate five methods of adding linguistic information to corpora that have been morphosyntactically annotated (=parsed) in the style of Penn treebanks. Some of these methods involve manual operations; some are executed by CorpusSearch functions; some require a combination of manual and automated procedures. Which method is used depends almost entirely on the type of information to be added and the goals of the user. Of course, the main goal, regardless of method, is to record within the corpus additional information that can be used for analysis and also retained through further searches and data processing.
This paper presents a case study of the use of the NINJAL Parsed Corpus of Modern Japanese (NPCMJ) for syntactic research. NPCMJ is the first phrase structure-based treebank for Japanese that is specifically designed for application in linguistic (in addition to NLP) research. After discussing some basic methodological issues pertaining to the use of treebanks for theoretical linguistics research, we introduce our case study on the status of the Coordinate Structure Constraint (CSC) in Japanese, showing that NPCMJ enables us to easily retrieve examples that support one of the key claims of Kubota and Lee (2015): that the CSC should be viewed as a pragmatic, rather than a syntactic constraint. The corpus-based study we conducted moreover revealed a previously unnoticed tendency that was highly relevant for further clarifying the principles governing the empirical data in question. We conclude the paper by briefly discussing some further methodological issues brought up by our case study pertaining to the relationship between linguistic research and corpus development.
This paper presents a case study of the use of the NINJAL Parsed Corpus of Modern Japanese (NPCMJ) for syntactic research. NPCMJ is the first phrase structure-based treebank for Japanese that is specifically designed for application in linguistic (in addition to NLP) research. After discussing some basic methodological issues pertaining to the use of treebanks for theoretical linguistics research, we introduce our case study on the status of the Coordinate Structure Constraint (CSC) in Japanese, showing that NPCMJ enables us to easily retrieve examples that support one of the key claims of Kubota and Lee (2015): that the CSC should be viewed as a pragmatic, rather than a syntactic constraint. The corpus-based study we conducted moreover revealed a previously unnoticed tendency that was highly relevant for further clarifying the principles governing the empirical data in question. We conclude the paper by briefly discussing some further methodological issues brought up by our case study pertaining to the relationship between linguistic research and corpus development.
The articles in this special issue are based on presentations made at the international symposium entitled Exploiting Parsed Corpora: Applications in Research, Pedagogy, and Processing held at the National Institute for Japanese Language and Linguistics (NINJAL) on Dec. 9-10, 2017 and organized by the collaborative research project at NINJAL entitled 'Development of and Linguistic Research with a Parsed Corpus of Japanese' with which all the guest editors are associated.
In this paper, we discuss constituent ordering generalizations in Japanese. Japanese has SOV as its basic order, but a significant range of argument order variations brought about by ‘scrambling’ is permitted. Although scrambling does not induce much in the way of semantic effects, it is conceivable that marked orders are derived from the unmarked order under some pragmatic or other motivations. The difference in the effect of basic and derived order is not reflected in native speaker’s grammaticality judgments, but we suggest that the intuition about the ordering of arguments may be attested in corpus data. By using the Keyaki treebank (a proper subset of which is NINJAL Parsed Corpus of Modern Japanese (NPCMJ)), it is shown that the naturally-occurring corpus data confirm that marked orderings of arguments are less frequent than their unmarked ordering counterparts. We suggest some possible motivations lying behind the argument order variations.
The principal barrier to the uptake of technologies in schools is not technological, but social and political. Teachers must be convinced of the pedagogical benefits of a particular curriculum before they will agree to learn the means to teach it. The teaching of formal grammar to first language students in schools is no exception to this rule. Over the last three decades, most schools in England have been legally required to teach grammatical subject knowledge, i.e. linguistic knowledge of grammar terms and structure, to children age five and upwards as part of the national curriculum in English. A mandatory set of curriculum specifications for England and Wales was published in 2014, and elsewhere similar requirements were imposed. However, few current English school teachers were taught grammar themselves, and the dominant view has long been in favour of ‘real books’ rather than the teaching of a formal grammar. English grammar teaching thus faces multiple challenges: to convince teachers of the value of grammar in their own teaching, to teach the teachers the knowledge they need, and to develop relevant resources to use in the classroom. Alongside subject knowledge, teachers need pedagogical knowledge – how to teach grammar effectively and how to integrate this teaching into other kinds of language learning. The paper introduces the Englicious web platform for schools, and summarises its development and impact since publication. Englicious draws data from the fully-parsed British Component of the International Corpus of English, ICE-GB. The corpus offers plentiful examples of genuine natural language, speech and writing, with context and potentially audio playback. However, corpus examples may be age-inappropriate or over-complex, and without grammar training, teachers are insufficiently equipped to use them. In the absence of grammatical knowledge among teachers, it is insufficient simply to give teachers and children access to a corpus. Whereas so-called ‘classroom concordancing’ approaches offer access to tools and encourage bottom-up learning, Englicious approaches the question of grammar teaching in a concept-driven, top-down way. It contains a modular series of professional development resources, lessons and exercises focused on each concept in turn, in which corpus examples are used extensively. Teachers must be able to discuss with a class why, for instance, work is a noun in a particular sentence, rather than merely report that it is. The paper describes the development of Englicious from secondary to primary, and outlines some of the practical challenges facing the design of this type of teaching resource. A key question, the ‘selection problem’, concerns how tools parameterise the selection of relevant examples for teaching purposes. Finally we discuss curricula for teaching teachers and the evaluation of the effectiveness of the intervention.
Abstract Meaning Representation (AMR) is a meaning representation framework in which the meaning of a full sentence is represented as a single-rooted, acyclic, directed graph. In this article, we describe an on-going project to build a Chinese AMR (CAMR) corpus, which currently includes 10,149 sentences from the newsgroup and weblog portion of the Chinese TreeBank (CTB). We describe the annotation specifications for the CAMR corpus, which follow the annotation principles of English AMR but make adaptations where needed to accommodate the linguistic facts of Chinese. The CAMR specifications also include a systematic treatment of sentence-internal discourse relations. One significant change we have made to the AMR annotation methodology is the inclusion of the alignment between word tokens in the sentence and the concepts/relations in the CAMR annotation to make it easier for automatic parsers to model the correspondence between a sentence and its meaning representation. We develop an annotation tool for CAMR, and the inter-agreement as measured by the Smatch score between the two annotators is 0.83, indicating reliable annotation. We also present some quantitative analysis of the CAMR corpus. 46.71% of the AMRs of the sentences are non-tree graphs. Moreover, the AMR of 88.95% of the sentences has concepts inferred from the context of the sentence but do not correspond to a specific word.
Previous work on light verb constructions (e.g. chorii kar ‘theft do; steal’) in Hindi describes their syntactic formation via co-predication (Ahmed et al., 2012, Butt, 2014). This implies that both noun and light verb contribute their arguments, and these overlapping argument structures must be composed in the syntax. In this paper, we present a co-predication analysis using Tree-Adjoining Grammar, which models syntactic composition and semantic selectional preferences without transformations (deletion or argument identification). The analysis has two key components (i) an underspecified category for the nominal and (ii) combinatorial constraints on the noun and light verb to specify selectional preferences. The former has the advantage of syntactic composition without argument identification and the latter prevents over-generalization, while recognizing the semantic contribution of both predicates. This work additionally accounts for the agreement facts for the Hindi LVC.
This paper contributes to the on-going discussion of how best to analyze and handle complex predicate formations, commenting in particular on the properties of Hindi N-V complex predicates as set out by Vaidya et al. (2019). I highlight features of existing LFG analyses and focus in particular on the modular architecture of LFG, its attendant multidimensional lexicon and the analytic consequences which follow from this. I point out where the previously existing LFG proposals have been misunderstood as viewed from the lens of theories such as LTAG and HPSG, which assume a very different architectural set-up and provide a comparative discussion of the issues.
This paper compares a recent TAG-based analysis of complex predicates in Hindi/Urdu with its HPSG analog. It points out that TAG combines actual structure while HPSG (and Categorial Grammar and other valence-based frameworks) specify valence of lexical items and hence potential structure. This makes it possible to have light verbs decide which arguments of embedded heads get realized, something that is not possible in TAG. TAG has to retreat to disjunctions instead. While this allows straight-forward analyses of active/passive alternations based on the light verb in valence-based frameworks, such an option does not exist for TAG and it has to be assumed that preverbs come with different sets of arguments.
Context-free grammars (CFG) were one of the first formal tools used to model natural languages, and they remain relevant today as the basis of several frameworks. A key ingredient of CFG is the presence of nested recursion. In this paper, we investigate experimentally the capability of several recurrent neural networks (RNNs) to learn nested recursion. More precisely, we measure an upper bound of their capability to do so, by simplifying the task to learning a generalized Dyck language, namely one composed of matching parentheses of various kinds. To do so, we present the RNNs with a set of random strings having a given maximum nesting depth and test its ability to predict the kind of closing parenthesis when facing deeper nested strings. We report mixed results: when generalizing to deeper nesting levels, the accuracy of standard RNNs is significantly higher than random, but still far from perfect. Additionally, we propose some non-standard stack-based models which can approach perfect accuracy, at the cost of robustness.
Previously unpublished analysis cited in "Lexical Factorization and Syntactic Behavior" in this volume.
We consider the extent to which different deep neural network (DNN) configurations can learn syntactic relations, by taking up Linzen et al.’s (2016) work on subject-verb agreement with LSTM RNNs. We test their methods on a much larger corpus than they used (a ⇠24 million example part of the WaCky corpus, instead of their ⇠1.35 million example corpus, both drawn from Wikipedia). We experiment with several different DNN architectures (LSTM RNNs, GRUs, and CNNs), and alternative parameter settings for these systems (vocabulary size, training to test ratio, number of layers, memory size, drop out rate, and lexical embedding dimension size). We also try out our own unsupervised DNN language model. Our results are broadly compatible with those that Linzen et al. report. However, we discovered some interesting, and in some cases, surprising features of DNNs and language models in their performance of the agreement learning task. In particular, we found that DNNs require large vocabularies to form substantive lexical embeddings in order to learn structural patterns. This finding has interesting consequences for our understanding of the way in which DNNs represent syntactic information. It suggests that DNNs learn syntactic patterns more efficiently through rich lexical embeddings, with semantic as well as syntactic cues, than from training on lexically impoverished strings that highlight structural patterns.
In this paper, we examine the correlation between lexical semantics and the syntactic realization of the different components of a word’s meaning in natural language. More specifically, we will explore the effect that lexical factorization in verb semantics has on the suppression or expression of semantic features within the sentence. Factorization was a common analytic tool employed in early generative linguistic approaches to lexical decomposition, and continues to play a role in contemporary semantics, in various guises and modified forms. Building on the unpublished analysis of verbs of seeing in Joshi (1972), we argue here that the significance of lexical factorization is twofold: first, current models of verb meaning owe much of their insight to factor-based theories of meaning; secondly, the factorization properties of a lexical item appear to influence, both directly and indirectly, the possible syntactic expressibility of arguments and adjuncts in sentence composition. We argue that this information can be used to compute what we call the factor expression likelihood (FEL) associated with a verb in a sentence. This is the likelihood that the overt syntactic expression of a factor will cooccur with the verb. This has consequences for the compositional mechanisms responsible for computing the meaning of the sentence, as well as significance in the creation of computational models attempting to capture linguistic behavior over large corpora.
This paper presents a new classification of verbs of change and modification, proposing a dynamic interpretation of the lexical semantics of the predicate and its arguments. Adopting the model of dynamic event structure proposed in Pustejovsky (2013), and extending the model of dynamic selection outlined in Pustejovsky and Jezek (2011), we define a verb class in terms of its Dynamic Argument Structure (DAS), a representation which encodes how the participants involved in the change behave as the event unfolds. We address how the logical resources and results of change predicates are realized syntactically, if at all, as well as how the exploitation of the resource results in the initiation or termination of a new object, i.e. the result. We show how DAS can be associated with a dynamically encoded event structure representation, which measures the change making reference to a scalar component, modelled in terms of assignment and/or testing of values of attributes of participants.
Abstract syntax is a semantic tree representation that lies between parse trees and logical forms. It abstracts away from word order and lexical items, but contains enough information to generate both surface strings and logical forms. Abstract syntax is commonly used in compilers as an intermediate between source and target languages. Grammatical Framework (GF) is a grammar formalism that generalizes the idea to natural languages, to capture cross-lingual generalizations and perform interlingual translation. As one of the main results, the GF Resource Grammar Library (GF-RGL) has implemented a shared abstract syntax for over 30 languages. Each language has its own set of concrete syntax rules (morphology and syntax), by which it can be generated from the abstract syntax and parsed into it. This paper presents a conversion method from abstract syntax trees to dependency trees. The method is applied for converting GF-RGL trees to Universal Dependencies (UD), which uses a common set of labels for different languages. The correspondence between GF-RGL and UD turns out to be good, and the relatively few discrepancies give rise to interesting questions about universality. The conversion also has potential for practical applications: (1) it makes the GF parser usable as a rule-based dependency parser; (2) it enables bootstrapping UD treebanks from GF treebanks; (3) it defines formal criteria to assess the informal annotation schemes of UD; (4) it gives a method to check the consistency of manually annotated UD trees with respect to the annotation schemes; (5) it makes information from UD treebanks available.
We investigate modality in Portuguese and we combine a linguistic perspective with an application-oriented perspective on modality. We design an annotation scheme reflecting theoretical linguistic concepts and apply this schema to a small corpus sample to show how the scheme deals with real world language usage. We present two schemas for Portuguese, one for spoken Brazilian Portuguese and one for written European Portuguese. Furthermore, we use the annotated data not only to study the linguistic phenomena of modality, but also to train a practical text mining tool to detect modality in text automatically. The modality tagger uses a machine learning classifier trained on automatically extracted features from a syntactic parser. As we only have a small annotated sample available, the tagger was evaluated on 11 modal verbs that are frequent in our corpus and that denote more than one modal meaning. Finally, we discuss several valuable insights into the complexity of the semantic concept of modality that derive from the process of manual annotation of the corpus and from the analysis of the results of the automatic labeling: ambiguity and the semantic and syntactic properties typically associated to one modal meaning in context, and also the interaction of modality with negation and focus. The knowledge gained from the manual annotation task leads us to propose a new unified scheme for modality that applies to the two Portuguese varieties and covers both written and spoken data.
