Phrase

About: Phrase is a research topic. Over the lifetime, 12580 publications have been published within this topic receiving 317823 citations. The topic is also known as: syntagma & phrases.

Call centre agent automated assistance

Abstract: This invention relates to the automated assistance of a telephony call center agent comprising a method and system for providing dialogue suggestions to an agent during an agent caller interaction. A prior art solution provides data based on analyzed text from one or other of the conversations but does not offer any useful information based on the performance of the agent or on the state of the interaction. A method of interaction in a voice response application comprising: receiving a voice signal in a call center; identifying whether the caller or a call center agent is the originator of the voice signal; converting the voice signal into computer readable text; identifying a key word such as a confrontational phrase e.g. ‘what are you talking about’ in the converted computer readable text; and providing a different suggestion depending on whether the originator is the call agent or the caller. For instance, a suggestion if the agent made the confrontational phrase would be to use a less confrontational phrase next time such as ‘can you explain that again’. A suggestion if the caller made the confrontational phrase would be to counter with a ‘I'll try to explain that better’.

Training Phrase Translation Models with Leaving-One-Out

Abstract: Several attempts have been made to learn phrase translation probabilities for phrase-based statistical machine translation that go beyond pure counting of phrases in word-aligned training data. Most approaches report problems with over-fitting. We describe a novel leaving-one-out approach to prevent over-fitting that allows us to train phrase models that show improved translation performance on the WMT08 Europarl German-English task. In contrast to most previous work where phrase models were trained separately from other models used in translation, we include all components such as single word lexica and reordering models in training. Using this consistent training of phrase models we are able to achieve improvements of up to 1.4 points in BLEU. As a side effect, the phrase table size is reduced by more than 80%.

On degree phrases and result clauses

Abstract: This dissertation consists of two parts, one on degree phrases and one on result clauses. Sentence (1) below exemplifies a result clause construction: (1) Zij had zo mooi gezongen dat het publiek er stil van was she has so beautifully sang that the audience there silent of was ‘She had sung so beautifully that the audience was still quiet afterwards’ Chapters 2 considers the status of a degree phrase adverbial in linguistic theory. The question addressed there is whether degree phrase adverbials are best analyzed as adjuncts or as specifiers. Though the structural status of adjuncts and specifiers is the same in the programme I adopt (cf. Kayne 1994, Hoekstra 1991), specifiers differ from other adjoined phrases in that they have an agreement relationship with the head of the projection they are adjoined to. The presence or absence of such an agreement relationship leads to different hypotheses and conclusions. After comparing the two viewpoints, I conclude that degree phrases functioning as modifiers are best analyzed as adjuncts. Having decided on the structural status of degree phrases that function as modifiers, I discus their internal structure in chapter 3. This discussion starts with an interesting set of data provided by Corver (1994, 1997). After presenting Corver’s interpretation of these data, in which he assumes a quantifier projection in between a degree head and the adjectival projection it selects, I argue that his interpretation is partially incorrect. Corver proposes to consider more, less, most and least as heads of the quantifier phrase. The same quantifier phrase would be present in a degree phrase like too beautiful, where it can be spelled out as much in examples of so-pronominalization: so much so. However, Corver’s assumptions yield a number of incorrect predictions. Instead, I show that more, less, most and least are composite forms, consisting of an adjectival quantifer like much or little and a degree head (the comparative morpheme -er or superlative morpheme -st). This implies that more etc. are full degree phrases, that may adjoin to (for instance) other degree phrases or prepositional phrases (cf. Doetjes, Neeleman & Van der Koot 1998 as well). Similarly, degree heads like so, too or as can combine with much to form a full degree phrase that may adjoin to a phrase it modifies. This explains Corver’s (1994, 1997) observation that these degree heads do not occur on their own in environments other than adjectival projections, whereas more etc. do allow for a number of other syntactic environments. Rather, more etc. has the same distribution as too much (instead of too by itself). Analyzing more as a degree phrase, just like too much is a degree phrase, neatly accounts for their similar distribution. I conclude that there is no quantifier projection that intervenes between a degree head and its complement, contra Corver’s (1994, 1997) analysis. Chapter 4 opens the second part of the dissertation, which concerns the analysis of the structural configuration of sentences with a result clause. In chapter 4 I present previous analyses of result clause constructions. I argue that a result clause is not generated inside the projection of the degree phrase it is associated with. Instead, I propose an analysis of result clause constructions in chapter 5 that generates the result clause in the (usually sentence-final) position that we observe. The analysis originates in the analysis (due to Jan Koster) that coordination and relative clauses show similar behaviour in several configurations. The observation that result clause constructions and comparatives also share these similarities provides firm support for the claim that these constructions all share the same structural configuration, viz. a conjunction phrase: (21) ConjP [...[ zo AP]...] ConjP degp Conj0 [dat...] : The conjunction phrase provides a structure that allows for a straightforward analysis of several properties that the four constructions at issue have in common. A result clause appears to be conjoined with the matrix clause or with a part of the matrix clause. That is to say, the specifier of the conjunction phrase in (2) can be the matrix clause or part of the matrix clause. The latter assumption is corroborated by topicalization data in which the conjunction phrase as a whole is topicalized. In addition, the conjunction analysis yields predictions with respect to the availability of verb-second main clause word order in Frisian result clauses (cf. chapter 6). Note that the availability of main clause properties in Frisian result clause is not straightforward in an extraposition analysis of result clauses: the availability of root properties in result clauses was noted to be an exception among adjunct clauses by for instance Zwart (1997a). The conjunction analysis of result clause constructions provides an account of why result clauses can have verb second word order, whereas adjunct clauses cannot: result clauses are not adjuncts, but conjuncts. Whenever a result clause is conjoined with the matrix clause, it can have verb second, which is the word order of root clauses. Whenever a result clause is arguably conjoined with part of the matrix clause, or with a subordinate clause, verb second word order is ungrammatical. Apparently, conjunction with a matrix clause provides the result clause with matrix clause status. This makes sense in a conjunction analysis, because two coordinated main clauses share the same configuration as a result clause that is conjoined with the matrix clause. In sum, the conjunction analysis of result clauses is very well supported. Additionally, the analysis obeys the constraints of Kayne’s (1994) program of the antisymmetry of syntax. Therefore, it also provides support for Kayne’s programme. Conclusion 1 there is no intermediary quantifier phrase in a degree phrase like te mooi ‘too beautiful’; 2 result clauses are conjoined with the matrix clause or with a part thereof.

Phrase units as determinants of visual processing in music reading.

Abstract: Keyboard musicians sight-read passages of music in which the amount of information about the presence of phrase units was systematically varied. A distinction was made between ‘physical’ unit markers, which allowed delineation of a unit prior to analysis of its component elements, and ‘structural’ unit markers, which defined a unit in terms of the sequential rules obeyed by its constituent elements. During the execution of the passages the text would be removed at a point known in advance only to the experimenter. Subjects were then required to execute all the material seen beyond this point to provide a measure of the eye-hand span at that point. It was found that the presence of structural markers increased span, and tended to cause span to extend exactly to a phrase boundary. The presence of physical markers did not increase span, although it also tended to cause span to extend exactly to a phrase boundary. The results suggest a clear analogy between the cognition of music and language, in that knowledge of abstract structure is of importance in the organization of immediate visual processing of text.