Identification Of Speakers In Novels - IT/Computer Science Assignment Help

Assignment Task

1 Introduction

Novels are important as social communication documents, in which novelists develop the plot by means of discourse between various characters. In spite of a frequently expressed opinion that all novels are simply variations of a certain number of basic plots (Tobias, 2012), every novel has a unique plot (or several plots) and a different set of characters. The interactions among characters, especially in the form of conversations, help the readers construct a mental model of the plot and the changing relationships between characters. Many of the complexities of interpersonal relationships, such as romantic interests, family ties, and rivalries, are conveyed by utterances

A precondition for understanding the relationship between characters and plot development in a novel is the identification of speakers behind all utterances. However, the majority of utterances are not explicitly tagged with speaker names, as is the case in stage plays and film scripts. In most cases, authors rely instead on the readers’ comprehension of the story and of the differences between characters.

Since manual annotation of novels is costly, a system for automatically determining speakers of utterances would facilitate other tasks related to the processing of literary texts. Speaker identification could also be applied on its own, for instance in generating high quality audio books without human lectors, where each character would be identifiable by a distinct way of speaking. In addition, research on spoken language processing for broadcast and multi-party meetings (Salamin et al., 2010; Favre et al., 2009) has demonstrated that the analysis of dialogues is useful for the study of social interactions.

In this paper, we investigate the task of speaker identification in novels. Departing from previous approaches, we develop a general system that can be trained on relatively small annotated data sets, and subsequently applied to other novels for which no annotation is available. Since every novel has its own set of characters, speaker identification cannot be formulated as a straightforward tagging problem with a universal set of fixed tags. Instead, we adopt a ranking approach, which enables our model to be applied to literary texts that are different from the ones it has been trained on.

2 Related Work

Previous work on speaker identification includes both rule-based and machine-learning approaches. Glass and Bangay (2007) propose a rule generalization method with a scoring scheme that focuses on the speech verbs. The verbs, such as said and cried, are extracted from the communication category of WordNet (Miller, 1995). The speech-verb-actor pattern is applied to the utterance, and the speaker is chosen from the available candidates on the basis of a scoring scheme. Sarmento and Nunes (2009) present a similar approach for extracting speech quotes from online news texts. They manually define 19 variations of frequent speaker patterns, and identify a total of 35 candidate speech verbs. The rule-based methods are typically characterized by low coverage, and are too brittle to be reliably applied to different domains and changing styles.

Elson and McKeown (2010) (henceforth referred to as EM2010) apply the supervised machine learning paradigm to a corpus of utterances extracted from novels. They construct a single feature vector for each pair of an utterance and a speaker candidate, and experiment with various WEKA classifiers and score-combination methods. To identify the speaker of a given utterance, they assume that all previous utterances are already correctly assigned to their speakers. Our approach differs in considering the utterances in a sequence, rather than independently from each other, and in removing the unrealistic assumption that the previous utterances are correctly identified.

The speaker identification task has also been investigated in other domains. Bethard et al. (2004) identify opinion holders by using semantic parsing techniques with additional linguistic features. Pouliquen et al. (2007) aim at detecting direct speech quotations in multilingual news. Krestel et al. (2008) automatically tag speech sentences in newspaper articles. Finally, Ruppenhofer et al. (2010) implement a rule-based system to enrich German cabinet protocols with automatic speaker attribution.

3 Definitions and Conventions

In this section, we introduce the terminology used in the remainder of the paper. Our definitions are different from those of EM2010 partly because we developed our method independently, and partly because we disagree with some of their choices. The examples are from Jane Austen’s Pride and Prejudice, which was the source of our development set.

An utterance is a connected text that can be attributed to a single speaker. Our task is to associate each utterance with a single speaker. Utterances that are attributable to more than one speaker are rare; in such cases, we accept correctly identifying one of the speakers as sufficient. In some cases, an utterance may include more than one quotationdelimited sequence of words, as in the following example.

In this case, the words said Jane are simply a speaker tag inserted into the middle of the quoted sentence. Unlike EM2010, we consider this a single utterance, rather than two separate ones.

We assume that all utterances within a paragraph can be attributed to a single speaker. This “one speaker per paragraph” property is rarely violated in novels — we identified only five such cases in Pride & Prejudice, usually involving one character citing another, or characters reading letters containing quotations. We consider this an acceptable simplification, much like assigning a single part of speech to each word in a corpus. We further assume that each utterance is contained within a single paragraph. Exceptions to this rule can be easily identified and resolved by detecting quotation marks and other typographical conventions.

4 Speaker Identification

In this section, we describe our method of extracting explicit speakers, and our ranking approach, which is designed to capture the speaker alternation pattern.

5 Features

In this section, we describe the set of features used in our ranking approach. The principal feature sets are listed in Table 2, together with an indication whether they are novel or have been used in previous work.

5.1 Basic Features

A subset of our features correspond to the features that were proposed by EM2010. These are mostly features related to speaker names. For example, since names of speakers are often mentioned in the vicinity of their utterances, we count the number of words separating the utterance and a name mention. However, unlike EM2010, we consider only the two nearest characters in each direction, to reflect the observation that speakers tend to be mentioned by name immediately before or after their corresponding utterances. Another feature is used to represent the number of appearances for speaker candidates. This feature reflects the relative importance of a given character in the novel. Finally, we use a feature to indicate the presence or absence of a candidate speaker’s name within the utterance. The intuition is that speakers are unlikely to mention their own name.

5.2 Vocatives

We propose a novel vocative feature, which encodes the character that is explicitly addressed in an utterance. For example, consider the following utterance:

Intuitively, the speaker of the utterance is neither Mr. Bingley nor Lizzy; however, the speaker of the next utterance is likely to be Lizzy. We aim at capturing this intuition by identifying the addressee of the utterance.

We manually annotated vocatives in about 900 utterances from the training set. About 25% of the names within utterance were tagged as vocatives. A Logistic Regression classifier (Agresti, 2006) was trained to identify the vocatives. The classifier features are shown in Table 3. The features are designed to capture punctuation context, as well as the presence of typical phrases that accompany vocatives. We also incorporate interjections like “oh!” and fixed phrases like “my dear”, which are strong indicators of vocatives. Under 10-fold cross validation, the model achieved an Fmeasure of 93.5% on the training set

This IT/Computer Science Assignment has been solved by our IT/Computer Science Experts at My Uni Paper. Our Assignment Writing Experts are efficient to provide a fresh solution to this question. We are serving more than 10000+ Students in Australia, UK & US by helping them to score HD in their academics. Our Experts are well trained to follow all marking rubrics & referencing style.

Be it a used or new solution, the quality of the work submitted by our assignment experts remains unhampered. You may continue to expect the same or even better quality with the used and new assignment solution files respectively. There’s one thing to be noticed that you could choose one between the two and acquire an HD either way. You could choose a new assignment solution file to get yourself an exclusive, plagiarism (with free Turnitin file), expert quality assignment or order an old solution file that was considered worthy of the highest distinction.