Search

Sentiment analysis of short texts is difficult for their simplicity and compactness. This goes a step further when it comes to the Chinese texts. Although deep learning achieved better accuracy in sentiment analysis, there is a lack of explain-ability. Thus, this paper evaluates the effectiveness of techniques for sentiment analysis of Chinese short financial texts with deep learning. For this, we built a Chinese short financial texts corpus (CSFC) and designed an ablation experiment. Beside the CFSC, we used a Chinese review collection and an English short-text repository in the experiment for comparison. There are five techniques involved. They are the Pinyin, the segmentation, the lexical analysis, the phrase structure and the attention mechanism. As results, we found that the phrase structure and the attention mechanism are two of the best. Therefore, the best model in the experiment is called a Phrase Structure and Attention-based Deep network model (PhraSAD). Moreover, to improve the classification accuracy on neutral data, we use a dual classifier strategy for 3-class problems. Experimental results showed that PhraSAD outperformed all other compared models on all experimental datasets.

A number of natural language processing tools for Urdu language processing have been developed in the past few years for word segmentation, part of speech tagging, chunking, named entity recognition and parsing. Corpora, especially treebanks, are essential data resources for language processing. This work presents the development and evaluation of an Urdu treebank, the CLE-UTB and a statistical parser. The treebank has been annotated with phrase structure annotation. Part of speech tagging has been performed semi-automatically by using an existing tagger and incorrect tags were corrected manually by annotators. The syntactic annotation has been performed in the Penn Treebank style to mark phrases. The annotation scheme also adds functional labels for grammatical roles. Currently, the treebank contains 7854 annotated sentences and 148,575 tokens. Completeness and correctness of the syntactic labels have been checked automatically after manual annotation. To ensure the annotation consistency of the resource, a grammar-based evaluation and an automatic consistency checking tool have been used to detect linguistically implausible constituents. The inter-annotator agreement is greater than 90%. We have developed a bidirectional long-short term memory (BiLSTM) based parser and a POS tagger which have been trained on the final version of the treebank. We have improved our results by training the word embeddings on a large Urdu text corpus. Our parser produced an f-score of 88.1% and the POS tagger performed with an accuracy of 96.3%.

Syntactic complexity has been recognized as an important construct in writing research, and for the past five decades, many syntactic complexity measures (SCMs) have been examined in numerous studies. This systematic review is the first study of its kind to synthesize 36 studies spanning from 1970 to 2019 by identifying and cataloging all SCMs examined during this period. An analysis was performed on how the use of SCMs varied by genre, grade level, students’ writing ability, and writing quality. Five online databases (Academic Search Premier, ERIC, PsycINFO, Psychology and Behavioral Sciences, and PsycARTICLES) were searched. A total of 48 SCMs were grouped into six categories: T-units and sentences, clauses, phrases, words, combined measures, and other measures. Most studies examined three common SCMs: mean T-unit length, mean number of words per clause, and mean number of clauses per T-unit. The argumentative genre had the highest values for T-unit length and mean number of clauses per T-unit, which could indicate greater sentence complexity. Higher-grade-level students generally construct sentences that are syntactically complex, but comparison between studies was difficult because different studies investigated different SCMs. Although students with higher writing abilities generally construct sentences that are syntactically more complex compared to students with lower writing abilities, the findings are not conclusive, as only a few studies examined this relationship. A similarly inconclusive relationship was found between syntactic complexity and writing quality because only a few studies examined this relationship. More research is needed to examine the relationship between SCMs, writing quality, and genre.

An elicited production study investigated subject-aux inversion in 5-year-old children with specific language impairment (SLI) and 2 control groups, typically-developing 5-year-old children and 3-year-old children matched by mean length of utterance. The experimental findings showed that children with specific language impairment produced subject-aux inversion in yes/no questions significantly less often than either of the control groups. However, the fact that lack of inversion is reflected in the input led to the proposal that children with specific language impairment choose the most economical grammatical option. For main clause wh-questions, children with SLI carried out subject-aux inversion at a rate that was not significantly different from the control groups. This finding suggests that these children have access to hierarchical phrase structure representations for questions and the relevant movement operations. In embedded wh-questions, where subject-aux inversion is not permitted, children with SLI implemented SAI more frequently than the control groups. Our interpretation of this finding is that once children with SLI acquire the subject-aux inversion rule, that they are slower to learn that embedded clauses present an exception to the rule.

It has proven hard to force a decision between rival analyses of Universal 20. This is because new typological data are scarce, and the number of syntactic tests applicable in the noun phrase is relatively small. I therefore consider a related set of facts that involve language-internal word order variation in the verb phrase. I first show that the pattern of grammatical and ungrammatical orders in Dutch verb phrases containing three PPs closely matches the pattern of attested and unattested orders in the noun phrase. I then use the distribution of the particle pas ‘only’ to argue that PP extraposition results from variation in the linearization of sister nodes. This means that the symmetric account of Universal 20 in Abels and Neeleman (Syntax 15:25–74, 2012) extends to the Dutch data, but the antisymmetric account in Cinque (Linguist Inquiry 36:315–332, 2005) does not.

Previous studies on Korean relative clauses (RC) show that, with respect to processing, object-extracted relative clauses (ORC) are more difficult to process at the head noun than subject-extracted relative clauses within temporarily ambiguous contexts. ORCs, however, are predicted by experience-based processing models to incur a greater processing cost during early processing stages at the RC verb, since it is a likely locus of disambiguation for RCs in Korean, and because ORCs are a less frequent structure. Consequently, the current study investigates whether processing difficulty for ORCs manifests itself at the RC verb using eye-tracking methods, a simple sentence structure and a sentential-decision task. The results revealed significantly increased go-past reading times for ORCs at the RC verb. We believe this is a result of a less frequent structure being more difficult to parse during disambiguation. Accordingly, experience-based models of processing can accurately predict difficulty for ORCs in Korean.

We report the results from two experiments examining native and non-native German speakers' sensitivity to crossover constraints on pronoun resolution. Our critical stimuli sentences contained personal pronouns in either strong (SCO) or weak crossover (WCO) configurations. Using eye-movement monitoring during reading and a gender-mismatch paradigm, Experiment 1 investigated whether a fronted wh-phrase would be considered as a potential antecedent for a pronoun intervening between the wh-phrase and its canonical position. Both native and non-native readers initially attempted coreference in WCO but not in SCO configurations, as evidenced by early gender-mismatch effects in our WCO conditions. Experiment 2 was an offline antecedent judgement task whose results mirrored the SCO/WCO asymmetry observed in our reading-time data. Taken together, our results show that the SCO constraint immediately restricts pronoun interpretation in both native and non-native comprehension, and further suggest that SCO and WCO constraints derive from different sources.

The current study tackles a long standing question of whether comprehenders perform structural revision when it is not forced by grammar or not. Using an eye-tracking reading paradigm, we addressed this issue by making use of global structural ambiguity in Japanese. Our results show that comprehenders initially associate a relative clause with the first potential head noun and that they revise this analysis when the second noun is lexico-semantically possible as the relative clause head, but do not when it is impossible. The results are incompatible with the Revision as Last Resort hypothesis. Instead, they support the parsing with unforced revision that is immediately sensitive to lexical properties. We argue that our results cannot be accounted for by serial modular processing models but that they can be explained by ranked-parallel interactive processing models. Furthermore, we propose that head-finality is a key factor involved in the availability of unforced revision.

We suggest that one way to approach the evolution of language is through reverse engineering: asking what components of the language faculty could have been useful in the absence of the full complement of components. We explore the possibilities offered by linear grammar, a form of language that lacks syntax and morphology altogether, and that structures its utterances through a direct mapping between semantics and phonology. A language with a linear grammar would have no syntactic categories or syntactic phrases, and therefore no syntactic recursion. It would also have no functional categories such as tense, agreement, and case inflection, and no derivational morphology. Such a language would still be capable of conveying certain semantic relations through word order-for instance by stipulating that agents should precede patients. However, many other semantic relations would have to be based on pragmatics and discourse context. We find evidence of linear grammar in a wide range of linguistic phenomena: pidgins, stages of late second language acquisition, home signs, village sign languages, language comprehension (even in fully syntactic languages), aphasia, and specific language impairment. We also find a full-blown language, Riau Indonesian, whose grammar is arguably close to a pure linear grammar. In addition, when subjects are asked to convey information through nonlinguistic gesture, their gestures make use of semantically based principles of linear ordering. Finally, some pockets of English grammar, notably compounds, can be characterized in terms of linear grammar. We conclude that linear grammar is a plausible evolutionary precursor of modern fully syntactic grammar, one that is still active in the human mind.

The current study investigated the distinction of L2 (second language) English article choice sensitivity in fifty-three L1-Korean L2-English learners in semantic contexts. In the context of English as a foreign language, the participants were divided into two groups based on grammatical ability as determined by their performance on a cloze test. In addition, a forced-choice elicitation test and a writing production test were administered to assess, respectively, the participants' receptive and productive article choice abilities. Regardless of grammatical ability, the results disclosed the overuse of the indefinite a in the [[Formula: see text]definite, -specific] context and the definite the in the [-definite, [Formula: see text]specific] context on the forced-choice elicitation test. In the [[Formula: see text]definite, [Formula: see text]specific] and [-definite, -specific] contexts, however, the overuse of either the indefinite a or the definite the, respectively, was less likely. Furthermore, it was revealed on the writing test that the participants more accurately used the definite the than the indefinite a, and they were also found to unreasonably omit more articles than to add or substitute articles on the writing production test. The findings across the two tests indicate that L1-Korean L2-English learners are more likely to have intrinsic difficulties transferring their L1 noun phrase (NP) knowledge to L2 NP knowledge owing to structural discrepancies and complex interfaces between L1 NPs and L2 NPs with respect to syntactic, semantic and pragmatic/discourse language subsystems.

The purpose of the present study was to compare the effects of input flooding and input elaboration on the learning of non-congruent collocations by Iranian English as a foreign language (EFL) learners. The participants of the study included 60 EFL learners at the intermediate level of language proficiency who were divided into two equal groups. They received the instruction of non-congruent collocations based on input elaboration and input flooding methods and their knowledge of non-congruent collocations was measured before and after the instruction. The results of paired samples t-tests run on the pretest and posttest scores of each group showed that input flooding and input elaboration had both significantly affected the learning of non-congruent collocations by the participants. However, the results of ANCOVA run on the pretest and posttest scores of non-congruent collocations test indicated that there was no significant difference between the two groups receiving input elaboration and input flooding in terms of their knowledge of non-congruent collocations after the treatment. This led to the conclusion that input flooding and input elaborations are two effective methods for teaching non-congruent collocations to Iranian EFL learners.

A number of previous studies on Chinese relative clauses (RC) have reported conflicting results on processing asymmetry. This study aims to revisit the prevalent debate on whether subject-extracted RCs (SRC) or object-extracted RCs (ORC) are easier to process by using the eye-movement technique. In the current study, the data are analyzed in terms of the gaze duration and regression of eye-movement in three critical areas: head noun, embedded verb, and RC-modifying noun phrase as subject. The results show an ORC preference for the processing of RC structures, which supports the word-order account and the Dependency Locality Theory, and a better cross-clausal integration for SRC, which supports the perspective-shift account. The processing asymmetry in Chinese RCs are discussed under relevant theoretical accounts, such as structure-based, memory-based, and perspective shift accounts. We argue that the findings are associated with the syntactic nature of Chinese (a head-initial language with pre-nominal RCs).

The present study aimed to investigate (a) the extent to which Chinese-speaking learners of English in Taiwan use referential noun phrase (NP) information contained in discourse contexts to complete ambiguous noun/verb fragments in a sentence completion task, and (b) whether and when they use the contexts to disambiguate main verb versus reduced relative clause (MV/RRC) ambiguities in real time. Results showed that unlike native English speakers, English learners did not create a marked increase in RRC completions in biasing two-NP-referent discourse contexts except for advanced learners. Nevertheless, like native speakers, the learners at elementary, intermediate, and advanced English proficiency levels all used the information in a later stage of resolving the MV/RRC ambiguities in real time. The delayed effect of referential context information observed suggests that L2 learners, like native speakers, are able to construct syntax-to-discourse mappings in real time. It also suggests that processing of syntactic information takes precedence over integration of syntactic information with discourse information during L1 and L2 online sentence processing.

Previous researches indicated that the performance of automatic prosodic boundaries prediction benefited from syntactic information for Mandarin. However, few researches focused on the usefulness of various syntactic features on different phrase boundaries prediction detailly, especially on large-scale corpus. This paper investigates the effect of different syntactic features combinations for prosodic phrase boundaries prediction based on a large-scale Mandarin corpus. The experimental results shows that prediction models of prosodic word and prosodic phrase achieve best performance with syntactic phrase and dependency features. Different from other two types of boundaries, the models with dependency features outperform than models with other features when prediction object is intonational phrase.

The development of precision grammars is an inherently resource-intensive process; their complexity means that changes made to one area of a grammar often introduce unexpected flow-on effects elsewhere in the grammar which may only be discovered after some time has been invested in updating numerous test suite items. In this paper, we present the browser-based gDelta tool, which aims to provide grammar engineers with more immediate feedback on the impact of changes made to a grammar by comparing parser output from two different grammar versions. We describe an attribute weighting algorithm for highlighting components of the grammar that have been strongly impacted by a modification to the grammar, as well as a technique for clustering test suite items whose parsability has changed, in order to locate related groups of effects. These two techniques are used to present the grammar engineer with different views on the grammar to inform them of different aspects of change in a data-driven manner.

A novel online reading methodology termed Self-Guided Reading (SGR) is examined to determine if it can successfully detect well-studied syntactic processing behaviours. In SGR, a participant runs their finger under masked text in order to reveal a sentence. It is therefore similar to self-paced reading in presentation of stimuli, but different in the motion that the participant makes to interact with the stimuli. The phenomena of relative clause, adverb and noun phrase/sentential attachment are utilised to allow comparison to previous research that employed self-paced reading and eye-tracking. SGR was able to detect the predicted processing behaviours in all sentence types. Moreover, once design choices and task effects are accounted for, SGR was the most consistent in triggering a motor movement change at the predicted point in the sentence. Able to provide a semi-continuous reading measure at low cost, SGR should be investigated further to uncover the full potential of the method for psycholinguistic research.

This study aimed at assessing the effects of letters’ connectivity in Arabic on visual word recognition. For this purpose, reaction times (RTs) and accuracy scores were collected from ninety-third, sixth and ninth grade native Arabic speakers during a lexical decision task, using fully connected (Cw), partially connected (PCw) and nonconnected (NCw) Arabic words and pseudowords. Effects of grade on word recognition (in RTs and accuracy) and word superiority were predicted to occur. Also, in the third grade, recognition of NCw was predicted to be faster and more accurate than recognition of Cw, because in previous studies NCw were assumed to be visually less complex. In sixth and ninth grades, due to the frequent exposure to connected forms, the recognition of Cw was predicted to be as fast as or faster and more accurate than NCw. The findings largely supported the first and the second hypotheses. As for the third graders, a mixed pattern was obtained, suggesting that the participants were probably in a transitional phase. The results from sixth graders clearly showed that NCw yielded the slowest response times and the lowest accuracy scores across connectivity conditions. Finally, for ninth graders, the absence of connectivity effects on the speed of processing was attributed to the use of very frequent and highly automatized words.

This study investigated 4-year-old children's knowledge of the structural constraints on ellipsis constructions in Mandarin Chinese, focusing on the structural difference between verb phrase (VP) ellipsis and null object constructions. In Mandarin Chinese, if the antecedent clause of a VP-ellipsis structure contains an adverbial modifier, this adverbial modifier must be recovered at the elided site in the second clause. However, if the antecedent clause of a null object construction contains an adverbial modifier, the reconstruction of this adverbial modifier at the elided site is not necessary. Two experiments were conducted to examine whether young Mandarin-speaking children are sensitive to this structural difference between the two ellipsis constructions. The results show that children exhibited adult-like performance in both experiments. This is evidence that young Mandarin-speaking children have knowledge of the structural difference between VP-ellipsis and null object constructions with regard to adverbial recovery. We interpreted the findings in conjunction with previous research as evidence that children's knowledge of the structural constraints on ellipsis constructions is innately specified.

This paper proposes an analysis of the Taiwanese DP justified by Taiwanese–Spanish code-switching data from highly competent bilingual speakers from a Taiwanese–Spanish community in Buenos Aires. In line with other research, we claim that Taiwanese has a NumP and an independent DP which cannot be filled by null elements. Further, we propose a new analysis for pre-classifier (Cl) adjectives (the tua ‘big’ and se ‘small’ type), which we show to be functional elements of what we call the Size Phrase. We also claim that Taiwanese has two classifier positions, the upper of which optionally appears for our speakers in constructions involving a Size element. In addition to our claims on the Taiwanese DP, our analysis provides evidence against claims that phrase structure is universal (Cinque, Adverbs and functional heads: a cross-linguistics perspective, 1999).

In this article we present a Spanish grammar implemented in the Linguistic Knowledge Builder system and grounded in the theoretical framework of Head-driven Phrase Structure Grammar. The grammar is being developed in an international multilingual context, the DELPH-IN Initiative, contributing to an open-source repository of software and linguistic resources for various Natural Language Processing applications. We will show how we have refined and extended a core grammar, derived from the LinGO Grammar Matrix, to achieve a broad-coverage grammar. The Spanish DELPH-IN grammar is the most comprehensive grammar for Spanish deep processing, and it is being deployed in the construction of a treebank for Spanish of 60,000 sentences based in a technical corpus in the framework of the European project METANET4U (Enhancing the European Linguistic Infrastructure, GA 270893GA; http://www.meta-net.eu/projects/METANET4U/ .) and a smaller treebank of about 15,000 sentences based in a corpus from the press.

I show that words with indefinite implicit complements occasion a dilemma for their model theory. There has been only two previous attempts to address this problem, one by Fodor and Fodor (1980) and one by Dowty (1981). Each requires that any word tolerating an implicit complement be treated as ambiguous between two different lexical entries and that a meaning postulate or lexical rule be given to constrain suitably the meanings of the various entries for the word. I show that the positing of such an ambiguity runs counter to the facts and propose an alternative solution which does not appeal to ambiguity, meaning postulates or lexical rules. Indeed, I show that the dilemma posed by indefinite implicit complements is posed by all implicit complements and that a general solution to the problem of implicit complements follows from an independently motivated, single treatment of five other problems, that of subcategorization, that of phrasal projections of words, that of defining a model theoretic structure for phrase structure grammars, that of complement polyvalence and that of complement polyadicity.

Comprehension includes interpreting sentences in terms of aspectual categories such as processes (Harry climbed) and culminations (Harry reached the top). Adding a verbal modifier such as for many years to a culmination coerces its interpretation from one to many culminations. Previous studies have found that coercion increases lexical decision and meaning judgment time, but not eye fixation time. This study recorded eye movements as participants read sentences in which a coercive adverb increased the interpretation of multiple events. Adverbs appeared at the end of a clause and line; the post-adverb region appeared at the beginning of the next line; follow-up questions occasionally asked about aspectual meaning; and clause type varied systematically. Coercive adverbs increased eye fixation time in the post-adverb region and in the adverb and post-adverb regions combined. Factors that influence the appearance of aspectual coercion may include world knowledge, follow-up questions, and the location and ambiguity of adverbs.

In this paper, we present our attempts to design and implement a large-coverage computational grammar for the Persian language based on the Generalized Phrase Structured Grammar (GPSG) model. This grammatical model was developed for continuous speech recognition (CSR) applications, but is suitable for other applications that need the syntactic analysis of Persian. In this work, we investigate various syntactic structures relevant to the modern Persian language, and then describe these structures according to a phrase structure model. Noun (N), Verb (V), Adjective (ADJ), Adverb (ADV), and Preposition (P) are considered basic syntactic categories, and X-bar theory is used to define Noun phrases, Verb phrases, Adjective phrases, Adverbial phrases, and Prepositional phrases. However, we have to extend Noun phrase levels in X-bar theory to four levels due to certain complexities in the structure of Noun phrases in the Persian language. A set of 120 grammatical rules for describing different phrase structures of Persian is extracted, and a few instances of the rules are presented in this paper. These rules cover the major syntactic structures of the modern Persian language. For evaluation, the obtained grammatical model is utilized in a bottom-up chart parser for parsing 100 Persian sentences. Our grammatical model can take 89 sentences into account. Incorporating this grammar in a Persian CSR system leads to a 31% reduction in word error rate.

We propose a layered-grammar model to represent actions. Using this model, an action is represented by a set of grammar rules. The bottom layer of an action instance's parse tree contains action primitives such as spatiotemporal (ST) interest points. At each layer above, we iteratively mine grammar rules and "super rules" that account for the high-order compositional feature structures. The grammar rules are categorized into three classes according to three different ST-relations of their action components, namely the strong relation, weak relation and stochastic relation. These ST-relations characterize different action styles (degree of stiffness), and they are pursued in terms of grammar rules for the purpose of action recognition. By adopting the Emerging Pattern (EP) mining algorithm for relation pursuit, the learned production rules are statistically significant and discriminative. Using the learned rules, the parse tree of an action video is constructed by combining a bottom-up rule detection step and a top-down ambiguous rule pruning step. An action instance is recognized based on the discriminative configurations generated by the production rules of its parse tree. Experiments confirm that by incorporating the high-order feature statistics, the proposed method largely improves the recognition performance over the bag-of-words models.

Phrase-based decoding is conceptually simple and straightforward to implement, at the cost of drastically oversimplified reordering models. Syntactically aware models make it possible to capture linguistically relevant relationships in order to improve word order, but they can be more complex to implement and optimise. In this paper, we explore a new middle ground between phrase-based and syntactically informed statistical MT, in the form of a model that supplements conventional, non-hierarchical phrase-based techniques with linguistically informed reordering based on syntactic dependency trees. The key idea is to exploit linguistically-informed hierchical structures only for those dependencies that cannot be captured within a single flat phrase. For very local dependencies we leverage the success of conventional phrase-based approaches, which provide a sequence of target-language words appropriately ordered and ready-made with any agreement morphology. Working with dependency trees rather than constituency trees allows us to take advantage of the flexibility of phrase-based systems to treat non-constituent fragments as phrases. We do impose a requirement--that the fragment be a novel sort of "dependency constituent"--on what can be translated as a phrase, but this is much weaker than the requirement that phrases be traditional linguistic constituents, which has often proven too restrictive in MT systems.

It is a commonly accepted view in the Hungarian linguistic literature that sentence structure is determined by information structure, viewed as a phrase structure theoretic interpretation of the topic–comment articulation of the sentence. There is a designated topic position at the left edge of the sentence, namely SpecTopP, hosting constituents that are claimed to be in a predicative relation with the rest of the sentence. On this view, topic–comment and logical subject–logical predicate are considered to be synonymous notions. We argue that the notion of topic as used in the Hungarian literature poses some serious problems, which can only be eliminated if the pragmatic aspects of topichood are separated from its semantic function entailing the development of a two-level approach to information structure. Topic and logical subject belong to two different levels with topic being an essentially pragmatic notion and logical subject being a syntactico-semantic notion. On this analysis the basic syntactic structure of the Hungarian sentence is determined by the articulation “logical subject–logical predicate” rather than by the articulation “topic-comment”. The proposed analysis has important typological consequences.

Two sentence processing experiments on a dative NP ambiguity in Korean demonstrate effects of phrase length on overt and implicit prosody. Both experiments controlled non-prosodic length factors by using long versus short proper names that occurred before the syntactically critical material. Experiment 1 found that long phrases induce different prosodic phrasing than short phrases in a read-aloud task and change the preferred interpretation of globally ambiguous sentences. It also showed that speakers who have been told of the ambiguity can provide significantly different prosody for the two interpretations, for both lengths. Experiment 2 verified that prosodic patterns found in first-pass pronunciations predict self-paced reading patterns for silent reading. The results extend the coverage of the Implicit Prosody Hypothesis [Fodor, J Psycholinguist Res 27:285-319, 1998; Prosodic disambiguation in silent reading. In M. Hirotani (Ed.), NELS 32 (pp. 113-132). Amherst, MA: GLSA Publications, 2002] to another construction and to Korean. They further indicate that strong syntactic biases can have rapid effects on the formulation of implicit prosody.

How the perceptions of novel figurative phrases change over the course of numerous presentations were addressed in three studies using rating tasks (Studies 1 and 3) and recall tasks (Study 2). The present set of studies replicated much of Schweigert et al.'s (J Psychol Res 32:455-475, 2003) findings of changes in correlations among figurative phrases ratings. The results also suggest that catchiness, frequency heard agreement, and frequency used agreement can be used as predictors of phrase memorability and potential predictors of later idiomaticity.

This study examined the relationship of different measures of syntactic complexity with rated quality for two genres of text produced by middle school students. It was hypothesized that different measures would be associated with distinct aspects of syntactic complexity; words per clause with greater use of structures more typical of expository texts, and clauses per T-unit with structures more typical of conversational or narrative registers. A sample of 41 seventh and eighth grade students from suburban middle schools composed a narrative and persuasive essays. Texts were rated for quality and coded for syntactic features including words per clause and clauses per T-unit. Syntactic complexity as measured in words per clause was positively correlated with quality for essays but not for narratives. Clauses per T-unit was positively correlated with quality for narratives, but negatively correlated with quality for essays. The relationships between syntactic complexity and text quality were thus found to be dependent both on the genre of the text, and the measure of syntactic complexity used.

We present a phrase-based statistical machine translation approach which uses linguistic analysis in the preprocessing phase. The linguistic analysis includes morphological transformation and syntactic transformation. Since the word-order problem is solved using syntactic transformation, there is no reordering in the decoding phase. For morphological transformation, we use hand-crafted transformational rules. For syntactic transformation, we propose a transformational model based on a probabilistic context-free grammar. This model is trained using a bilingual corpus and a broad-coverage parser of the source language. This approach is applicable to language pairs in which the target language is poor in resources. We considered translation from English to Vietnamese and from English to French. Our experiments showed significant BLEU-score improvements in comparison with Pharaoh, a state-of-the-art phrase-based SMT system.

What is common to all languages is notation, so Universal Grammar can be understood as a system of notational types. Given that infants acquire language, it can be assumed to arise from some a priori mental structure. Viewing language as having the two layers of calculus and protocol, we can set aside the communicative habits of speakers. Accordingly, an analysis of notation results in the three types of Identifier, Modifier and Connective. Modifiers are further interpreted as Quantifiers and Qualifiers. The resulting four notational types constitute the categories of Universal Grammar. Its ontology is argued to consist in the underlying cognitive schema of Essence, Quantity, Quality and Relation. The four categories of Universal Grammar are structured as polysemous fields and are each constituted as a radial network centred on some root concept which, however, need not be lexicalized. The branches spread out along troponymic vectors and together map out all possible lexemes. The notational typology of Universal Grammar is applied in a linguistic analysis of the ‘parts of speech’ using the English language. The analysis constitutes a ‘proof of concept’ in (1) showing how the schema of Universal Grammar is capable of classifying the so-called ‘parts of speech’, (2) presenting a coherent analysis of the verb, and (3) showing how the underlying cognitive schema allows for a sub-classification of the auxiliaries.

Noam Chomsky's work on ‘generative grammar’ led to the concept of a set of rules that can generate a natural language with a hierarchical grammar, and the idea that this represents a uniquely human ability. In a series of experiments with European starlings, in which several types of ‘warble’ and ‘rattle’ took the place of words in a human language, the birds learnt to classify phrase structure grammars in a way that met the same criteria. Their performance can be said to be almost human on this yardstick. So if there are language processing capabilities that are uniquely human, they may be more context-free or at a higher level in the Chomsky hierarchy. Or perhaps there is no single property or processing capacity that differentiates human language from non-human communication systems. Humans regularly produce new utterances that are understood by other members of the same language community1. Linguistic theories account for this ability through the use of syntactic rules (or generative grammars) that describe the acceptable structure of utterances2. The recursive, hierarchical embedding of language units (for example, words or phrases within shorter sentences) that is part of the ability to construct new utterances minimally requires a ‘context-free’ grammar2,3 that is more complex than the ‘finite-state’ grammars thought sufficient to specify the structure of all non-human communication signals. Recent hypotheses make the central claim that the capacity for syntactic recursion forms the computational core of a uniquely human language faculty4,5. Here we show that European starlings (Sturnus vulgaris) accurately recognize acoustic patterns defined by a recursive, self-embedding, context-free grammar. They are also able to classify new patterns defined by the grammar and reliably exclude agrammatical patterns. Thus, the capacity to classify sequences from recursive, centre-embedded grammars is not uniquely human. This finding opens a new range of complex syntactic processing mechanisms to physiological investigation.

Structure analysis of table form documents is an important issue because a printed document and even an electronic document do not provide logical structural information but merely geometrical layout and lexical information. To handle these documents automatically, logical structure information is necessary. In this paper, we first analyze the elements of the form documents from a communication point of view and retrieve the grammatical elements that appear in them. Then, we present a document structure grammar which governs the logical structure of the form documents. Finally, we propose a structure analysis system of the table form documents based on the grammar. By using grammar notation, we can easily modify and keep it consistent, as the rules are relatively simple. Another advantage of using grammar notation is that it can be used for generating documents only from logical structure. In our system, documents are assumed to be composed of a set of boxes and they are classified as seven box types. Then the box relations between the indication box and its associated entry box are analyzed based on the semantic and geometric knowledge defined in the document structure grammar. Experimental results have shown that the system successfully analyzed several kinds of table forms.

Hudson (1990) proposes that each conjunct in a coordinate phrase forms dependency relations with heads or dependents outside the coordinate phrase (the "multi-head" view). This proposal is tested through corpus analysis of Wall Street Journal text. For right-branching constituents (such as direct-object NPs), a short-long preference for conjunct ordering is observed; this is predicted by the multi-head view, under the assumption that structures resulting in shorter dependencies are preferred. A short-long preference is also observed for left-branching constituents (such as subject NPs), which is less obviously accommodated by the multi-head view but not incompatible with it. The repetition of determiners was also examined (the dog and cat versus the dog and the cat), and a stronger preference was found for repetition with singular count nouns as opposed to mass or plural nouns; this accords well with the multi-head view, under the reasoning that single-determiner constructions require crossing dependencies with count nouns but not with plural or mass nouns.

This work introduces a general phrase recognition system based on perceptrons, and a global online learning algorithm to train them together. The method applies to complex domains in which some structure has to be recognized. This global problem is broken down into two layers of local subproblems: a filtering layer, which reduces the search space by identifying plausible phrase candidates; and a ranking layer, which builds the optimal phrase structure by discriminating among competing phrases. A recognition-based feedback rule is presented which reflects to each local function its committed errors from a global point of view, and allows to train them together online as perceptrons. As a result, the learned functions automatically behave as filters and rankers, rather than binary classifiers, which we argue to be better for this type of problems. Extensive experimentation on partial parsing tasks gives state-of-the-art results and evinces the advantages of the global training method over optimizing each function locally and independently.

In this paper, I propose an implementation of relative clause extraposition in German. The proposal builds on Kiss (Natural Language and Linguistic theory, in press) who treats relative clause extraposition as an anaphoric process by means of percolation of anchors to which the relative clause is bound. I discuss several sources of spurious ambiguity in Kiss’s original formulation and suggest a two-step percolation of anchors that crucially distinguishes right-peripheral from central or left-peripheral percolation. Since extraposition is fairly productive, and phrase-structure alternates between head initial (prepositional phrases, V-initial) and head-final structures (postpositional phrases, V-final), German provides a good testing ground for techniques controlling spurious ambiguity that may easily be ported to languages where phrase structure is more canonical and/or extraposition more restricted. Finally, the performance of the Kiss-style approach is compared to an alternative implementation in terms of rightward movement, similar to Keller (1995, Towards as Account of Extraposition in HPSG. In Proceedings of the ninth Meeting of the European ACL. Dublin).

The LinGO Redwoods initiative is a seed activity in the design and development of a new type of treebank. A treebank is a (typically hand-built) collection of natural language utterances and associated linguistic analyses; typical treebanks—as for example the widely recognized Penn Treebank (Marcus, Santorini, & Marcinkiewicz, 1993), the Prague Dependency Treebank (Hajic, 1998), or the German TiGer Corpus (Skut, Krenn, Brants, & Uszkoreit, 1997)—assign syntactic phrase structure or tectogrammatical dependency trees over sentences taken from a naturallyoccuring source, often newspaper text. Applications of existing treebanks fall into two broad categories: (i) use of an annotated corpus in empirical linguistics as a source of structured language data and distributional patterns and (ii) use of the treebank for the acquisition (e.g. using stochastic or machine learning approaches) and evaluation of parsing systems. While several mediumto large-scale treebanks exist for English (and some for other major languages), all pre-existing publicly available resources exhibit the following limitations: (i) the depth of linguistic information recorded in these treebanks is comparatively shallow, (ii) the design and format of linguistic representation in the treebank hard-wires a small, predefined range of ways in which information can be extracted from the treebank, and (iii) representations in existing treebanks are static and over the (often yearor decade-long) evolution of a large-scale treebank tend to fall behind theoretical advances in formal linguistics and grammatical representation. LinGO Redwoods aims at the development of a novel treebanking methodology, (i) rich in nature and dynamic in both (ii) the ways linguistic data can be retrieved from the treebank in varying granularity and (iii) the constant evolution and regular updating of the treebank itself, synchronized to the development of ideas in syntactic theory. Starting in October 2001, the project is aiming to build the foundations for this new type of treebank, develop a basic set of tools required for treebank construction and maintenance, and construct an initial set of 10,000 annotated trees to be distributed together with the tools under an open-source license. Building a largescale treebank, disseminating it, and positioning the corpus as a widely-accepted resource is a multi-year effort; the results of this seeding activity will serve as a proof of concept for the novel approach that is expected to enable the LinGO group at CSLI both to disseminate the approach to the wider academic and industrial audience and to secure appropriate funding for the realization and exploitation of a larger treebank. The purpose of publication at this early stage is three-fold: (i) to encourage feedback on the Redwoods approach from a broader academic audience, (ii) to facilitate exchange with related work at other sites, and (iii) to invite additional collaborators to contribute to the construction of the Redwoods treebank or start its exploitation as early-access versions become available. This paper is an updated version of an earlier project report published by Oepen, Callahan, Flickinger, and Manning (2002); changes over that version include more recent numbers on the current Redwoods development status, inclusion of an example of discriminator-based disambiguation, and minor adaptations and corrections in various parts of the discussion.

This paper looks at integrating dependency and constituency into acommon framework, using the TAG formalism and a differentperspective on the meta-level grammar of Dras (1999a) in which themeta level models dependencies and the object level modelsconstituency. This framework gives consistent dependency analysesof raising verbs interacting with bridge verbs, additionally givinga solution to a problem in Synchronous TAG, and gives appropriateanalyses of subject-auxiliary inversion. This and other evidencesuggests the integration of dependency and constituency is a usefulavenue to explore.

Positron emission tomography (PET) was used to localize the brain regions involved in the processing of pauses and intonation changes, which underlie the syntactically correct perception of auditory verbal stimuli. Subjects were asked to listen to a phrase and to choose a correct answer from two variants presented on a monitor screen. Differences in cerebral blood circulation were mapped for perception of phrases containing or lacking a pause determining the meaning. Conscious analysis of the phrase structure was associated with activation of the right lower prefrontal area and the right posterior medial area of the cerebellum. The possible role of these brain structures in analyzing factors of syntagmatic splitting is discussed.

English gerunds such as (We weretalking about) John having a sabbaticalcombine the internal characteristics ofa clause with the external characteristicsof a noun phrase. Previous analyses havetried to recognise the mixed characterof gerunds by assigning them two separatenodes, one verbal and the other nominal.However, dependency analyses such as WordGrammar allow only one node perword, so they do not allow analyses of thiskind. Two-node analyses are strongevidence against dependency analysis, so itis important to be sure that they areneeded. The present paper presents ananalysis similar to the one recently proposedby Malouf in which the verbal and nominalclassifications are combined on asingle node which inherits both verbaland nominal characteristics; butunlike Malouf's analysis, it does notassume phrase structure. Like his, it exploitsthe logic of multiple default inheritance whichallows a single node to inherit fromtwo supercategories, in this case from both`verb' and `noun'. As Malouf points out,multiple inheritance works because English grammaris organised in such a way thatthe characteristics of these categoriesare orthogonal. In short, a gerund is both a verband a noun, as in traditional analyses. Simplestipulations are needed to allow for`possessive' subjects (e.g., about John'shaving a sabbatical) and a number ofvery specific constructions peculiar to gerunds:no in prohibitions or existentials(e.g., No playing loud music! There's no mistaking that voice), and a very fewconstructions which demand a gerund ratherthan a noun phrase (e.g., It's nouse . . ., They prevented us from ...).

The presence of temporal adjectives in possessive nominals like John's former car creates two interpretations. On one reading, the temporal adjective modifies the common noun (N-modifying reading). On the other, it modifies the possession relation (POSS-modifying reading). An explanation for this behavior is offered that appeals to what occurs in possessive sentences like John has a former car (N-modifying reading) and John formerly had a car (POSS-modifying reading). In the sentential cases, the source of two readings is two distinct, modifiable phrases. Given the parallels, we propose a structure for possessive nominals analogous to that of possessive clauses. Specifically, we argue that such nominals include a locative small-clause structure, following Freeze (1992), and we explain the ambiguity structurally, as a simple matter of where temporal adjectives attach (NP vs. PP). We show that this analysis provides a straightforward basis for the semantic composition of possessive nominals.

In many 'free' word order languages, it is not uncommon to findfixed word order phenomena in which a certain canonical word orderbecomes fixed under special circumstances. This phenomenon is termedword order freezing. The central dynamic in word order freezing is hierarchy alignment. In Hindi and Korean, for instance, word orderis free when the unmarked association among grammatical functions,semantic roles, case and positions in phrase structure matches therelative prominence relations of these dimensions. However, free wordorder becomes fixed when there is more than onemarked association of elements in these different dimensions of prominencein a single clause. This preference for avoiding highly markedassociations of prominence hierarchies in word order is a case of themore general phenomenon 'markedness reduction' in typologically markedgrammatical contexts. This paper develops an approach to word ordervariation in two scrambling languages, Hindi and Korean, withinOptimality Theory that is capable of subsuming both the free orderingand fixed ordering of constituents under the universal theory ofmarkedness. It accounts in a uniform manner for the universal basis ofword order freezing, while at the same time allowing for the range ofcrosslinguistic and language-internal variation that is observed.

This paper examines Ga/No Conversion in Japanese under the Move F theory of movement (Chomsky (1995)). Building on Miyagawa's (1993) analysis, the present paper argues that a genitive phrase raises out of a prenominal gapless clause in either overt or covert syntax. This claim is crucially based on Lasnik's (1999a) analysis of Exceptional Case Marking (ECM) constructions in English, according to which the ECM subject raises into a higher clause either overtly or covertly. It is also demonstrated that when a genitive subject originates in a relative clause, its raising is limited to covert syntax. This non-uniform behavior of the genitive subject is argued to follow from a particular implementation of Attract F.

It has been observed that phonological phrasing in Shanghai distinguishes certain determiners from others and wh-quantifier phrases from non-wh-quantifier phrases (Jin (1986), cited in Hung (1987) and Selkirk and Shen (1990)). In this paper, I show that such phonological phrasing distinctions are also found in Hokkien but in a more restricted environment. I propose a unified analysis of the phrasing facts in Shanghai and Hokkien. Specifically, I claim that the phonological phrasing rules are sensitive to the feature [±definite]. A DP specified for the feature [−definite] is phrased differently from one that is specified for [+definite] or unspecified with respect to the definiteness feature. In addition, I claim that phonological phrasing rules cannot see a syntactic projection with no phonological content. I propose the following phonological phrasing parameters for Shanghai and Hokkien Chinese. a.Shanghai Chinese Phrasing Parameter: Mark the left edge of all overt XPs, except [−definite] XPs. b.Hokkien Chinese Phrasing Paremeter: Mark the right edge of all overt XPs, except [−definite] XPs.

German sentences which were either correct, contained a selectional restriction violation, or a word category violation were presented auditorily to 16 native speakers of German (L1 group) and to 16 native speakers of Russian, who had learned German after the age of 10 (L2 group). Semantic violations elicited an N400 effect for both groups, but with a reduced amplitude and a longer peak latency in the L2 group. Compared to correct sentences, sentences with a phrase structure violation elicited an early anterior negativity followed by a broad centro-parietal positivity in native speakers. By contrast, there was no differential modulation of the early anterior negativity in the L2 group. A late positivity was also elicited in the second language learners, but it was slightly delayed compared to that shown by native speakers. This pattern is discussed in terms of different degrees of automaticity with respect to the subprocesses involved in sentence comprehension.

As is well known, in conditionals and, more generally, in structures involving adverbs of quantification, indefinite NPs like a cat display a variable quantificational force. Within DRT this phenomenon is analyzed by assimilating indefinites to variables. Unlike other variable-like elements, however, indefinites cannot be anaphoric to something else. That is, one cannot say things like "a cat usually meows if a cat is hungry" meaning "a cat usually meows if it is hungry." This is generally explained in terms of a novelty condition: indefinites must introduce novel variables. Cheng and Huang (1996) discuss and analyze two types of Chinese conditionals in which wh- words display quantificational variability. In one type of conditional, their behavior is fully analogous to that of indefinites. In the other, they behave like indefinites in the antecedent, while in the consequent they must be interpreted as bound pronouns. Thus, in DR-theoretic terms, Chinese wh-words obey the novelty condition in the antecedent but not in the consequent of a conditional. This behavior is unexpected. The present paper addresses this issue. The main claim is that a certain version of Dynamic Semantics leads one to expect elements with exactly the properties of Chinese wh-words. In particular, Dynamic Semantics makes it possible to reverse, in a sense, the classic DR-theoretic strategy. One can view indefinites as existentially quantified terms: however, their existential force can be overridden by operators in their local environment that wipe out their existential force, as it were, and get them to act like variables. If one takes this line, the Novelty Condition becomes dispensable and the problem disappears. The behavior of Chinese wh-words is also compared to that of other elements analyzable as indefinite pronominals, such as si in Italian or one in English.

In this paper I develop an argument for a pied-piping approach to the apparent absence of island effects in Japanese, along the lines of Nishigauchi (1986, 1990). The argument (first mentioned in Watanabe (1992b), crediting Mamoru Saito) has to do with the behavior of multiple wh-phrases in situ which are in an island; such wh-phrases must all take the same scope. On a covert pied-piping approach to island effects, the ban on distinct scopes follows straightforwardly; the island must be pied-piped to a scope position by the wh-phrases inside it, establishing the scope position for all of them. I show that the ban on distinct-scope readings does exhibit several properties of islands, including additional-wh effects. I then go on to investigate briefly the nature of pied-piping, developing a theory which accounts for the fact that wh-islands cannot be pied-piped.

Both semantic and syntactic context constraints can influence word processing at the level of lexical integration. In event-related brain potentials (ERPs), semantic integration is reflected by a negativity around 400 msec (N400), whereas phrase structure assignment and syntactic integration are assumed to be reflected by an early left anterior negativity and a late positivity (P600), respectively. An ERP study is presented in which participants read different types of sentences whose terminal verb was either congruent with the preceding context or incongruent due to a phrase structure violation, a semantic violation, or both. The main finding was that only the pure semantic violation condition, but not the combined semantic and syntactic violation condition, elicited a large N400. The two conditions containing phrase structure violations were predominantly characterized by a P600. Both semantic violation conditions, moreover, displayed a late negativity around 700 msec that overlapped with the P600 in the double violation condition. The absence of an N400 effect for elements that are syntactically as well as semantically incongruent with prior context suggests an early influence of phrase structure information on processes of lexical-semantic integration. The present data are discussed in comparison to previous ERP findings, and a new view of lexical integration processes is proposed.

This article proposes that, with Fiengo and May's (1994) indexing theory, a full-fledged reconstruction approach to null argument phenomena in such languages as Chinese, Japanese, and Korean becomes possible. Not only does the present NP-Ellipsis analysis preserve the insight of Otani and Whitman (1991), but it also helps overcome the drawbacks of their VP-Ellipsis analysis of null objects. This article thus argues that null NPs in these languages must be empty phrase- markers underlyingly, not empty pronouns per se, and hence, their reference properties can be determined only at LF after reconstruction. It is shown that this analysis also promises to solve certain binding-theoretic puzzles that null argument phenomena have posed.

This paper examines the so-called internally headed relative clause construction in Japanese, with particular focus on new data that involve quantificational NPs and wh-phrases. The data provide arguments for representations in which the internal head remains internal at LF. Furthermore, it is shown that the interpretation of this construction involves E-type anaphora, providing evidence for Hoshi's (1995) suggestion. An explicit mechanism for compositional interpretation is proposed, which also derives a restriction on possible internal heads. This study has the cross-linguistic implication that the constructions called "internally headed relative clauses" in various languages do not form a homogeneous class.