lynMarkdigRenderTest11 1.0.0

Tasks

<script>
  /**
   * Writes the current year to all elements that match the selector.
   */
  function setCurrentYear() {
    const YEAR_ELEMENTS_SELECTOR = '[fs-hacks-element="year"]';

    const yearElements = document.querySelectorAll(YEAR_ELEMENTS_SELECTOR);
    const currentYear = new Date().getFullYear().toString();

    yearElements.forEach((yearElement) => {
      yearElement.textContent = currentYear;
    });
  }
  
  document.addEventListener('DOMContentLoaded', setCurrentYear);
</script>

Word Sense Disambiguation Task

The task of choosing the correct sense for a word is called word sense disambiguation (WSD). WSD algorithms take an input word w in its context with a fixed set of potential word senses S_w of that input word and produce an output chosen from S_w. In the isolated WSD task, one usually uses the set of senses from a dictionary or theasurus like WordNet.

In the literature, there are actually two variants of the generic WSD task. In the lexical sample task, a small selected set of target words is chosen, along with a set of senses for each target word. For each target word w, a number of corpus sentences (context sentences) are manually labeled with the correct sense of w. In all-words task, systems are given entire sentences and a lexicon with the set of senses for each word in each sentence. Annotators are then asked to disambiguate every word in the text.

In all-words WSD, a classifier is trained to label the words in the text with their set of potential word senses. After giving the sense labels to the words in our training data, the next step is to select a group of features to discriminate different senses for each input word.

Named Entity Recognition Task

In named entity recognition, one tries to find the strings within a text that correspond to proper names (excluding TIME and MONEY) and classify the type of entity denoted by these strings. The problem is difficult partly due to the ambiguity in sentence segmentation; one needs to extract which words belong to a named entity, and which not. Another difficulty occurs when some word may be used as a name of either a person, an organization or a location. For example, Deniz may be used as the name of a person, or - within a compound - it can refer to a location Marmara Denizi 'Marmara Sea', or an organization Deniz Taşımacılık 'Deniz Transportation'.

The standard approach for NER is a word-by-word classification, where the classifier is trained to label the words in the text with tags that indicate the presence of particular kinds of named entities. After giving the class labels (named entity tags) to our training data, the next step is to select a group of features to discriminate different named entities for each input word.

[_ORG Türk Hava Yolları] bu [_TIME Pazartesi'den] itibaren [_LOC İstanbul] [_LOC Ankara] hattı için indirimli satışlarını [_MONEY 90 TL'den] başlatacağını açıkladı.

[_ORG Turkish Airlines] announced that from this [_TIME Monday] on it will start its discounted fares of [_MONEY 90TL] for [_LOC İstanbul] [_LOC Ankara] route.

Shallow Parse Task

Many language processing tasks do not require complex parse trees. Instead, a partial parse, or a shallow parse of a sentence is sufficient. Shallow parsing is the process of identifying flat non-overlapping parts of a sentence. These parts typically include Özne, Yüklem, Nesne, Zarf Tümleci, and Dolaylı Tümleç. Since a parsed text does not include a hierarchical structure, a bracketing notation is sufficient to denote the location and the type of shallow parse chunks in a sentence.

In shallow parsing, one tries to find the strings of text that belong to a chunk and to classify the type of that chunk. Standard approach for shallow parsing is a word-by-word classification, where the classifier is trained to label the words in the text with tags that indicate the presence of particular chunks. After giving the class labels to our training data chunk labels, the next step is to select a group of features to discriminate different chunks for each input word.

[_OZNE Türk Hava Yolları] [_{ZARF TÜMLECİ} Salı günü] [_NESNE yeni indirimli fiyatlarını] [_YÜKLEM açıkladı]

[_SUBJECT Turkish Airlines] [_PREDICATE announced] [_OBJECT new discounted fares] [_{ADVERBIAL CLAUSE} on Tuesday]

Semantic Role Labeling Task

Semantic Role Labeling (SRL) is a well-defined task where the objective is to analyze propositions expressed by the verb. In SRL, each word that bears a semantic role in the sentence has to be identified. There are different types of arguments (also called ’thematic roles’) such as Agent, Patient, Instrument, and also of adjuncts, such as Locative, Temporal, Manner, and Cause. These arguments and adjuncts represent entities participating in the event and give information about the event characteristics.

In the field of SRL, PropBank is one of the studies widely recognized by the computational linguistics communities. PropBank is the bank of propositions where predicate- argument information of the corpora is annotated, and the semantic roles or arguments that each verb can take are posited.

Each verb has a frame file, which contains arguments applicable to that verb. Frame files may include more than one roleset with respect to the senses of the given verb. In the roleset of a verb sense, argument labels Arg0 to Arg5 are described according to the meaning of the verb. For the example below, the predicate is “announce” from PropBank, Arg0 is “announcer”, Arg1 is “entity announced”, and ArgM- TMP is “time attribute”.

[_ARG0 Türk Hava Yolları] [_ARG1 indirimli satışlarını] [_ARGM-TMP bu Pazartesi] [_PREDICATE açıkladı].

[_ARG0 Turkish Airlines] [_PREDICATE announced] [_ARG1 its discounted fares] [_ARGM-TMP this Monday].

Detailed Description

• AnnotatedDataSetGenerator
• InstanceGenerator

AnnotatedDataSetGenerator

DataSet yaratmak için AnnotatedDataSetGenerator sınıfı önce üretilir.

AnnotatedDataSetGenerator(string directory, string pattern, InstanceGenerator instanceGenerator)

Ardından Generate metodu ile DataSet yaratılır.

DataSet Generate()

InstanceGenerator

DataGeneratorlerin InstanceGeneratorlere ihtiyacı vardır. Bunlar bir tek kelimeden bir Instance yaratan sınıflardır.

Instance GenerateInstanceFromSentence(Sentence sentence, int wordIndex)

NER problemi için NerInstanceGenerator, FeaturedNerInstanceGenerator ve VectorizedNerInstanceGeneratorsınıfı

ShallowParse problemi için ShallowParseInstanceGenerator, FeaturedShallowParseInstanceGenerator ve VectorizedShallowParseInstanceGenerator sınıfı

WSD problemi için SemanticInstanceGenerator, FeaturedSemanticInstanceGenerator ve VectorizedSemanticInstanceGenerator sınıfı

Morphological Disambiguation problemi için FeaturedDisambiguationInstanceGenerator sınıfı

Cite

If you use this resource on your research, please cite the following paper:

@article{acikgoz,
  title={All-words word sense disambiguation for {T}urkish},
  author={O. Açıkg{\"o}z and A. T. G{\"u}rkan and B. Ertopçu and O. Topsakal and B. {\"O}zenç and A. B. Kanburoğlu and {\.{I}}. Çam and B. Avar and G. Ercan and O. T. Y{\i}ld{\i}z},
  journal={2017 International Conference on Computer Science and Engineering (UBMK)},
  year={2017},
  pages={490-495}
}
@inproceedings{ertopcu17,  
	author={B. {Ertopçu} and A. B. {Kanburoğlu} and O. {Topsakal} and O. {Açıkgöz} and A. T. {Gürkan} and B. {Özenç} and İ. {Çam} and B. {Avar} and G. {Ercan} and O. T. {Yıldız}},  
	booktitle={2017 International Conference on Computer Science and Engineering (UBMK)},  title={A new approach for named entity recognition},   
	year={2017},  
	pages={474-479}
}