Underspecified Representations Research Articles

Ontology-Based Interpretation of Natural Language Philipp Cimiano, Christina Unger, and John McCrae (University of Arminia Bielefeld, Germany) Morgan & Claypool, Synthesis Lectures on Human Language Technologies, March 2014, 178 pages, (doi:10.2200/S00561ED1V01Y201401HLT024) , $45.00

<b>Ontology-Based Interpretation of Natural Language</b> <b>Philipp Cimiano, Christina Unger, and John McCrae</b> (University of Arminia Bielefeld, Germany) Morgan &amp; Claypool, Synthesis Lectures on Human Language Technologies, March 2014, 178 pages, (doi:10.2200/S00561ED1V01Y201401HLT024) , $45.00

Opposing effects of semantic diversity in lexical and semantic relatedness decisions.

Semantic ambiguity has often been divided into 2 forms: homonymy, referring to words with 2 unrelated interpretations (e.g., bark), and polysemy, referring to words associated with a number of varying but semantically linked uses (e.g., twist). Typically, polysemous words are thought of as having a fixed number of discrete definitions, or “senses,” with each use of the word corresponding to one of its senses. In this study, we investigated an alternative conception of polysemy, based on the idea that polysemous variation in meaning is a continuous, graded phenomenon that occurs as a function of contextual variation in word usage. We quantified this contextual variation using semantic diversity (SemD), a corpus-based measure of the degree to which a particular word is used in a diverse set of linguistic contexts. In line with other approaches to polysemy, we found a reaction time (RT) advantage for high SemD words in lexical decision, which occurred for words of both high and low imageability. When participants made semantic relatedness decisions to word pairs, however, responses were slower to high SemD pairs, irrespective of whether these were related or unrelated. Again, this result emerged irrespective of the imageability of the word. The latter result diverges from previous findings using homonyms, in which ambiguity effects have only been found for related word pairs. We argue that participants were slower to respond to high SemD words because their high contextual variability resulted in noisy, underspecified semantic representations that were more difficult to compare with one another. We demonstrated this principle in a connectionist computational model that was trained to activate distributed semantic representations from orthographic inputs. Greater variability in the orthography-to-semantic mappings of high SemD words resulted in a lower degree of similarity for related pairs of this type. At the same time, the representations of high SemD unrelated pairs were less distinct from one another. In addition, the model demonstrated more rapid semantic activation for high SemD words, thought to underpin the processing advantage in lexical decision. These results support the view that polysemous variation in word meaning can be conceptualized in terms of graded variation in distributed semantic representations.

The receptive-expressive gap in bilingual children with and without primary language impairment.

In this study, the authors examined the magnitude of the discrepancy between standardized measures of receptive and expressive semantic knowledge, known as a receptive-expressive gap, for bilingual children with and without primary language impairment (PLI). Spanish and English measures of semantic knowledge were administered to 37 Spanish-English bilingual 7- to 10-year old children with PLI and to 37 Spanish-English bilingual peers with typical development (TD). Parents and teachers completed questionnaires that yielded day-by-day and hour-by-hour information regarding children's exposure to and use of Spanish and English. Children with PLI had significantly larger discrepancies between receptive and expressive semantics standard scores than their bilingual peers with TD. The receptive-expressive gap for children with PLI was predicted by current English experience, whereas the best predictor for children with TD was cumulative English experience. As a preliminary explanation, underspecified phonological representations due to bilingual children's divided language input as well as differences in their languages' phonological systems may result in a discrepancy between standardized measures of receptive and expressive semantic knowledge. This discrepancy is greater for bilingual children with PLI because of the additional difficulty these children have in processing phonetic information. Future research is required to understand these underlying processes.

Making sense of Kafka: Structural biases induce early sense commitment for metonyms

Abstract Prior research suggests that the language processor initially activates an underspecified representation of a metonym consistent with all its senses, potentially selecting a specific sense if supported by contextual and lexical information. We explored whether a structural heuristic, the Subject as Agent Principle, which provisionally assigns an agent theta role to canonical subjects, would prompt immediate sense selection. In Experiment 1, we found initial evidence that this principle is active during offline and online processing of metonymic names like Kafka. Reading time results from Experiments 2 and 3 demonstrated that previous context biasing towards the metonymic sense of the name reduced, but did not remove, the agent preference, consistent with Frazier’s (1999) proposal that the processor may avoid selecting a specific sense, unless grammatically required.

Inequality across consonantal contrasts in speech perception: Evidence from mismatch negativity.

The precise structure of speech sound representations is still a matter of debate. In the present neurobiological study, we compared predictions about differential sensitivity to speech contrasts between models that assume full specification of all phonological information in the mental lexicon with those assuming sparse representations (only contrastive or otherwise not predictable information is stored). In a passive oddball paradigm, we studied the contrast sensitivity as reflected in the mismatch negativity (MMN) response to changes in the manner of articulation, as well as place of articulation of consonants in intervocalic positions of nonwords (manner of articulation: [edi ~ eni], [ezi ~ eni]; place of articulation: [edi ~ egi]). Models that assume full specification of all phonological information in the mental lexicon posit equal MMNs within each contrast (symmetric MMNs), that is, changes from standard [edi] to deviant [eni] elicit a similar MMN response as changes from standard [eni] to deviant [edi]. In contrast, models that assume sparse representations predict that only the [ezi] ~ [eni] reversals will evoke symmetric MMNs because of their conflicting fully specified manner features. Asymmetric MMNs are predicted, however, for the reversals of [edi] ~ [eni] and [edi] ~ [egi] because either a manner or place property in each pair is not fully specified in the mental lexicon. Our results show a pattern of symmetric and asymmetric MMNs that is in line with predictions of the featurally underspecified lexicon model that assumes sparse phonological representations. We conclude that the brain refers to underspecified phonological representations during speech perception.

Individual constraints for information structure

This paper, in the context of multilingual MT, proposes the use of ICONS (Individual CONstraintS) to add a representation of information structure to MRS. The value of ICONS is a list of objects of type info-str, each of which has the features CLAUSE and TARGET. The subtypes of info-str indicate which information structural role is played by the TARGET with respect to the CLAUSE. This proposal is designed to support both the calculation of focus projection from underspecified representations and the handling of multiclausal sentences.

Leveraging underspecification in knowledge artifacts to foster collaborative activities in professional communities

Collaborative problem solving often involves actors with heterogeneous competences or that see a common problem from different perspectives: this can make mutual understanding difficult. The paper presents case studies in different domains where collaboration leverages shared representations, and discusses the main reasons why these representations succeeded in fostering mutual understanding. We observed how the technologies proposed to manage those representations were successful only to the extent they were made able to adapt to the dynamic and open conventions that actors adopted during their activities. The point of the paper is that locality, openness and underspecification are key factors in this process, for their capability to promote tacit knowledge and to let competent actors reach a sufficient level of mutual understanding towards some common goal. The paper proposes a conceptual framework to characterize the notion of knowledge artifact interpreted as a semiotic system where actors can make sense of shared and underspecified representations, and derives from this notion implications for the design of a supportive technology.

Neuronal spoken word recognition: The time course of processing variation in the speech signal

Recent neurobiological studies revealed evidence for lexical representations that are not specified for the coronal place of articulation (PLACE; Friedrich, Eulitz, & Lahiri, 2006; Friedrich, Lahiri, & Eulitz, 2008). Here we tested when these types of underspecified representations influence neuronal speech recognition. In a unimodal auditory–auditory word fragment priming experiment target words either had initial coronal PLACE (e.g., tiger) or had initial noncoronal PLACE (e.g., pony). Prime-target pairs were either identical (e.g., ti-tiger, po-pony), differed in initial PLACE (e.g., pi-tiger, to-pony), or were unrelated. Event-related Potentials for identical and related targets started to differ from unrelated targets at 100 ms, indicating early perceptual processing. Different effects for coronal vs. noncoronal variation were observed starting at 200 ms. Results are related to recent neurocognitive research on compensation for assimilation of PLACE (e.g., rainbow realised as *raimbow). A time line of processing PLACE variation at multiple stages is proposed and parallel processing of specified and underspecified representations at the lexical level is discussed.

Are French dyslexic children sensitive to consonant sonority in segmentation strategies? Preliminary evidence from a letter detection task

This paper aims to investigate whether--and how--consonant sonority (obstruent vs. sonorant) and status (coda vs. onset) within syllable boundaries modulate the syllable-based segmentation strategies. Here, it is questioned whether French dyslexic children, who experience acoustic-phonetic (i.e., voicing) and phonological impairments, are sensitive to an optimal 'sonorant coda-obstruent onset' sonority profile as a cue for a syllable-based segmentation. To examine these questions, we used a modified version of the illusory conjunction paradigm with French dyslexic children compared with both chronological age-matched and reading level-matched controls. Our results first showed that the syllable-based segmentation is developmentally constrained in visual identification: in normally reading children, it appears to progressively increase as reading skills increase. However, surprisingly, our results also showed that dyslexic children were able to use syllable-sized units. Then, data highlighted that a syllable-based segmentation in visual identification basically relies on an optimal 'sonorant coda-obstruent onset' sonority profile rather than on phonological and orthographic statistical properties in normally reading children as well as, surprisingly, in dyslexic children. Our results are discussed to support a sonority-modulated prelexical role of syllable-sized units in visual identification in French, even in dyslexic children who exhibited a developmentally delayed profile. We argue that dyslexic children have deficits in online phonetic-phonological processing rather than degraded or underspecified phonetic-phonological representations.

“What you encode is not necessarily what you store”: Evidence for sparse feature representations from mismatch negativity

The present study examines whether vowels embedded in complex stimuli may possess underspecified representations in the mental lexicon. The second goal was to assess the possible interference of the lexical status of stimuli under study. Minimal pairs of German nouns differing only in the stressed vowels [e], [ø], and [o], and derived pseudowords, were used to measure the Mismatch Negativity (MMN) in a passive oddball-paradigm. The differing vowels were chosen such that the place of articulation information was conflicting vs. non-conflicting in the framework of models assuming underspecified representations in the mental lexicon (i.e. minimizing featural information by omitting redundant information in order to ensure efficient speech processing), whereas models assuming fully specified phonological representations would predict equal levels of conflict in all possible contrasts. The observed pattern of MMN amplitude differences was in accordance to predictions of models assuming underspecified phonological representations. As the possible interferences by other levels of linguistic processing was demonstrated, it seems favorable to use pseudowords for investigating phonological effects by means of MMN.

Investigations on markedness, syncretism and zero exponence in morphology

The inventories of subject clitics in northern Italian varieties are affected by widespread morphological changes. One of these changes is syncretism. In addition, many dialects display defective paradigms where some of the clitics are missing. The clitics that undergo syncretism and the clitics that are missing tend to be the same: the first person singular and plural, the second plural, and, less frequently, the third plural. Working in the framework of Distributed Morphology, I argue that these clitics are the target of markedness constraints that trigger repair operations leading to either syncretism or the removal of forms from a paradigm. In particular, I propose two such operations: one is feature deletion, which deletes a marked feature specification. This operation is always followed by feature insertion, which inserts the opposite value of the deleted feature specification (this operation leads to syncretism). The other one is obliteration, which deletes a morphosyntactic node and leads to underspecified morphosyntactic representations (this operation leads to paradigmatic gaps). I investigate how these operations affect the morphological exponents in PF, creating syncretisms and gaps in exponence.

The Linguistic Affiliation Constraint and phoneme recognition in diglossic Arabic

This study tested the effect of the phoneme's linguistic affiliation (Standard Arabic versus Spoken Arabic) on phoneme recognition among five-year-old Arabic native speaking kindergarteners (N=60). Using a picture selection task of words beginning with the same phoneme, and through careful manipulation of the phonological properties of target phonemes and distractors, the study showed that children's recognition of Standard phonemes was poorer than that of Spoken phonemes. This finding was interpreted as indicating a deficiency in the phonological representations of Standard words. Next, the study tested two hypotheses regarding the specific consequences of under-specified phonological representations: phonological encoding versus phonological processing. These hypotheses were addressed through an analysis of the relative power of distractors. The findings revealed that children's difficulty in accessing Standard Arabic phonemes was due to a difficulty in the phonological encoding of Standard words. We discuss the implications of the findings for language and literacy development in diglossic Arabic.

Indeterminacy, complex features and underspecification

At its simplest, morphosyntactic agreement may be viewed as involving linguistic objects which have the same values for a given feature. In contemporary constraint-based formalisms the relationship is usually modelled by structure sharing in the syntax; for predicate-argument agreement, most often it is assumed that the target and the controller provide compatible (complete or partial) specifications of features of the controller. This approach has much to recommend it, being based on the straightforward mechanism of combining information monotonically in the syntax. Additionally, it is often (but not invariably) assumed that (a subset of) these same morphosyntactic features are simply passed down to the morphology for spell out (Bresnan, In: Baltin and Collins (eds.) Handbook of contemporary syntactic theory, 2000; Lexical functional syntax, 2001) (but see e.g. Ackerman and Webelhuth (In: Butt and King (eds.) Online proceedings of the LFG96 conference, 1996); Sadler and Spencer (Yearbook of morphology 2002:71–97, 2001); Sells (In: Sadler and Spencer (eds.) Projecting morphology, 2004; In: Nikolaeva (ed.) Finiteness: theoretical and empirical foundations, 2007); Sadler and Nordlinger (In: Sadler and Spencer (eds.) Projecting morphology, 2004); Hinrichs and Nakazawa (Proceedings of HPSG2001, 2002) for approaches in lfg and other constraint-based formalisms which allow some further separation between these components). We discuss a range of cases of morphosyntactic agreement which are challenging in one way or another to this lean approach and explore how they may be modelled without adding any extra machinery, by using underspecified feature representations in the syntax, within one such framework, that of Lexical Functional Grammar lfg.

On the alternation of the modesty suffix in Middle Korean

This paper reanalyzes the phonological alternation of the 15 th century Middle Korean pre-final modesty suffix showing the various alternating forms, - s? p-,-z? p-, -c? p-,-s? β-, -z? β-, -c? β-,-s?m-, -z?m-, -c?m-, s’? β-, -z’? β-, -c’? β-, etc. As this suffix is no longer used in Modern Korean, there have been very few accounts on the phonological variations, compared with those on syntactic characteristics (H.-K. Kim 1947, Heo 1953, B.-H. Ahn 1961, C.-W. Park 1996, Y.-P. Kim 2002, etc.). There remains, however, a controversy over the underlying representation of this suffix in that those two segments /?/ and /β/ are extinct in the Modern Korean phonemic inventory. Based on this historical background, this paper attempts to analyze the phonological nature of this suffix from a new perspective. To this end, we propose an underspecified feature geometric representation of this pre-final suffix, reflecting the ambivalence of the initial and the final consonants. And this proposal for the underspecified representation is supported by various alternating forms. Finally, we conclude this paper by illustrating how the Middle Korean form has been changed to the Early Modern Korean variants in terms of the constraint ranking relationship and the change of hierarchy. Here we employ the combination tableau format (McCarthy 2008) in order to show the clear ranking relationship among the constraints.

Semantic Underspecification

AbstractSemantic underspecification is a technique to capture several readings of an ambiguous expression in one single representation by deliberately omitting the differences between the readings in the representation. First, underspecification formalisms will be presented to introduce underspecification in general and to outline important properties of these formalisms that allow their classification into subgroups. After expounding the kinds of ambiguity to which underspecification can be applied, the article then presents various motivations for the use of underspecification, and shows how underspecified semantic representations can be further processed.

The neural underpinnings of semantic ambiguity and anaphora

We used event-related brain potentials (ERPs) in order to investigate how definite NP anaphors are integrated into semantically ambiguous contexts. Although sentences such as Every kid climbed a tree lack any syntactic or lexical ambiguity, these structures exhibit two possible meanings, where either many trees or only one tree was climbed. This semantic ambiguity is the result of quantifier scope ambiguity. Previous behavioural studies have shown that a plural definite NP continuation is preferred (as reflected in a continuation sentence, e.g., The trees were in the park) over singular NPs (e.g., The tree was in the park). This study aimed to identify the neurophysiological pattern associated with the integration of the continuation sentences, as well as the time course of this process. We examined ERPs elicited by the noun and verb in continuation sentences following ambiguous and unambiguous context sentences. A sustained negative shift was most evident at the Verb position in sentences exhibiting scope ambiguity. Furthermore, this waveform did not differentiate itself until 900 ms after the presentation of the Noun, suggesting that the parser waits to assign meaning in contexts exhibiting quantifier scope ambiguity, such that such contexts are left as underspecified representations.

MMN and mapping the fine structure of phonological representations

Event Abstract Back to Event MMN and mapping the fine structure of phonological representations C. Eulitz1* 1 Department of Linguistics, University of Konstanz, Germany The Mismatch Negativity (MMN) has been used among others to map language specific representations in the brain. This presentation will summarize MMN studies dealing with the processing of the sound structure of languages. The focus will be on studies investigating the fine structure of phonological representations in the mental lexicon. I will argue that the MMN is a useful measure to compare the predictive value of different models of speech perception. Models proposing abstract underspecified representations of phonological features in the mental lexicon (i.e. not all phonological features are stored) predict asymmetric MMN effects which could be demonstrated for different featural dimensions and stimuli of different phonetic complexity. Constraints in using the MMN for investigations concerning phonological representations will be discussed. These results as well as studies demonstrating an influence of the language-specific phonological system on the fine structure of the phonological representations (i.e. identical and productive acoustic/phonetic contrasts may result in differential MMN effects in speakers of different languages) suggest an influence of use-dependent and language system-dependent factors on the fine structure of phonological representations in the mental lexicon. Conference: MMN 09 Fifth Conference on Mismatch Negativity (MMN) and its Clinical and Scientific Applications, Budapest, Hungary, 4 Apr - 7 Apr, 2009. Presentation Type: Oral Presentation Topic: Symposium 5: Language and language deficits Citation: Eulitz C (2009). MMN and mapping the fine structure of phonological representations. Conference Abstract: MMN 09 Fifth Conference on Mismatch Negativity (MMN) and its Clinical and Scientific Applications. doi: 10.3389/conf.neuro.09.2009.05.023 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 19 Mar 2009; Published Online: 19 Mar 2009. * Correspondence: C. Eulitz, Department of Linguistics, University of Konstanz, Konstanz, Germany, .carsten.eulitz@uni-konstanz.de Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers C. Eulitz Google C. Eulitz Google Scholar C. Eulitz PubMed C. Eulitz Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.

Scope and Situation Binding in LTAG Using Semantic Unification

This paper sets up a framework for Lexicalized Tree Adjoining Grammar (LTAG) semantics that brings together ideas from different recent approaches addressing some shortcomings of LTAG semantics based on the derivation tree. The approach assigns underspecified semantic representations and semantic feature structure descriptions to elementary trees. Semantic computation is guided by the derivation tree and consists of adding feature value equations to the descriptions. A rigorous formal definition of the framework is given. Then, within this framework, an analysis is proposed that accounts for the different scopal properties of quantificational NPs (including nested NPs), adverbs, raising verbs and attitude verbs. Furthermore, by integrating situation variables in the semantics, different situation binding possibilities are derived for different types of quantificational elements.

Neurophysiological evidence for underspecified lexical representations: Asymmetries with word initial variations.

How does the mental lexicon cope with phonetic variants in recognition of spoken words? Using a lexical decision task with and without fragment priming, the authors compared the processing of German words and pseudowords that differed only in the place of articulation of the initial consonant (place). Across both experiments, event-related brain potentials indicated that pseudowords with initial noncoronal place (e.g., *Brachen) activate words with initial coronal place (e.g., Drachen [dragon]). In contrast, coronal pseudowords (e.g., *Drenze) do not as effectively activate noncoronal words (e.g., Grenze [border]). Thus, certain word onset variations do not hamper the speech recognition system. The authors interpret this asymmetry as a consequence of underspecified coronal place of articulation in the mental lexicon.

Underspecification of syntactic ambiguities: Evidence from self-paced reading

Syntactically ambiguous sentences are sometimes read faster than disambiguated strings. Models of parsing have explained this tendency by appealing either to a race in the construction of alternative structures or to reanalysis. However, it is also possible that readers of ambiguous sentences save time by strategically underspecifying interpretations of ambiguous attachments. In a self-paced reading study, participants viewed sentences with relative clauses that could attach to one of two sites. Type of question was also manipulated between participants in order to test whether goals can influence reading/parsing strategies. The experiment revealed an ambiguity advantage in reading times, but only when participants expected superficial comprehension questions. When participants expected queries about relative clause interpretation, disambiguating regions were inspected with more care, and the ambiguity advantage was attenuated. However, even when participants expected relative clause queries, question-answering times suggested underspecified representations of ambiguous relative clause attachments. The results support the construal and "good-enough" models of parsing.