Critical Journal Report

Lecturer : Dra. Mei Suri., MA

Course : Semantics and Pragmatics

Arranged By :

Group 4

Harry Try Ananda Simatupang

Kartika Putri Anjani Lubis

Indah Suryani Ompusunggu

M. Rangga Ananda Ritonga

Rizka Putri Damanik

English Literature Department

Art and Language Faculty
State University of Medan
Medan
2018
 Preface
The task of this assignment is to find related journal article about regarding chapter
(Entailment) and determine its scope (lexical entailment).

Finally, this review is expected to encourage the readers. This review is also contain
the mistaken explanations which is the lack of the reviewer’s comprehension. Critizing and
giving reccomendations are also recommended to the reviewers in purpose to enhance the
reviewer ability in reviewing in the next even.

Medan, March 2018

Team
I. Identity of the Journals

1. Title : HyperLex: A Large-Scale Evaluation of Graded Lexical

Entailment

Author(s) : Ivan Vulic, Daniela Gerz, Douew Keila, Felix Hill and
Anna Korhonen

Publication Year : 2017

Publisher : The MIT Press Journal

2. Title : Towards a Probabilistic Model for Lexical Entailment

Author: Eyal Shnarch
Computer Science Department
Bar-Ilan University
Ramat-Gan, Israel
shey@cs.biu.ac.il
Jacob Goldberger
School of Engineering
Bar-Ilan University
Ramat-Gan, Israel
goldbej@eng.biu.ac.il
Ido Dagan
Computer Science Department
Bar-Ilan University
Ramat-Gan, Israel
dagan@cs.biu.ac.il

3. Title of the journal : A Lexical Alignment Model For Probablistic Textual

Alignment
Author : Oren Glickman, Ido Dagan and Moshe Koppel

4. Tittle : Entailment above the word level in distributional semantics

Author : Marco Baroni

Raffaela Bernardi
Ngoc-Quynh Do
Chung-chieh Shan

Pages : 10 Pages
II. Summary

1. HyperLex is a data set and evaluation resource that quantifies the extent of the
semantic category membership, that is, type-of relation, also known as
hyponymy–hypernymy or lexical entailment (LE) relation between 2,616 concept
pairs.

Most native speakers of English, in almost all contexts and situations, would
agree that dogs, cows, or cats are animals, and that tables or pencils are not.
However, for certain concepts, membership of the animal category is less clear-
cut. Whether lexical concepts such as dinosaur, human being, or amoeba are
considered animals seems to depend on the context in which such concepts are
described, the perspective of the speaker or listener, and even the formal scientific
knowledge of the interlocutors.

The quick overview consist of six aspects.

1. Natural Language Processinng

2. Representation Learning
3. Data Mining: Extending Knowledge Bases
4. Cognitive Science
5. Information Search
6. Beyond The Horizon: Multi-Model Answering

2. The development of HyperLex was principally inspired and motivated by

several factors. First, unlike prior work on lexical entailment in NLP, it focuses on
the relation of graded or soft lexical entailment at a continuous scale: The relation
quantifies the strength of the type-of relation between concepts rather than simply
making a binary decision as with the ungraded LE variant. Graded LE is firmly
grounded in cognitive linguistic theory of class prototypes and graded membership
stating that some concepts are more central to a broader category/class than others
(prototypicality) or that some concepts are only within the category to some extent
(graded membership). For instance, basketball is more frequently cited as a
prototypical sport than is chess or wrestling. One purpose of HyperLex is to
examine the effects of prototypicality and graded membership in human
judgments, as well as to provide a large repository (i.e., HyperLex contains 2,616
word pairs in total) of concept pairs annotated for graded lexical entailment.

The author makes this journal to know modeling entailment at lexical-level.

entailment systems are widely used. Therefore, people needed to know the lexical
in entailment. Textual Entailment was proposed as a generic paradigm for applied
semantic inference (Dagan et al., 2006. The author derives sentence-level
entailment decision base on lexical-level entailment evidence. Typically, this is
done by quantifying the degree of lexical coverage of the hypothesis terms by the
text terms (where a term may be multi-word). A hypothesis term is covered by a
text term if either they are identical (possibly at the stem or lemma level) or there is
a lexical entailment rule suggesting the entailment of the former by the latter.

The author aim at obtaining a probabilistic score for the likelihood that the
hypothesis terms are entailed by the terms of the text. There are several prominent
aspects of entailment, mostly neglected by previous lexical methods, which our
model aims to capture: (1) the reliability variability of different lexical resources;
(2) the effect of the length of transitive rule application chain on the likelihood of
its validity; and (3) addressing cases of multiple entailment evidence when
entailing a term.

The author analysis the data with specify the process by which a decision of
lexical entailment between T and H using knowledge resources should be
determined, specifies the probability of a particular chain c, connecting a text term
t to a hypothesis term h, to correspond to a valid entailment between t and h. And
after that, gives the probability that T entails all of H (T → H), assuming
independence of H’s .

There are two ways by which a term h ∈ H is entailed by a term

t ∈ T . A direct MATCH is the case in which t and h are identical terms (possibly at
the stem or lemma level). Alternatively, lexical entailment can be established based
on knowledge of entailing lexical-12 semantic relations, such as synonyms,
hypernyms and morphological derivations, available in lexical resources. These
relations provide lexical entailment rules, e.g. Jaguar → car. We denote the
resource which provided the rule r by R(r). It should be noticed at this point that
such rules specify a lexical entailment relation that might hold for some (T, H)
pairs but not necessarily for all pairs, e.g. the rule Jaguar → car does not hold in
the wildlife context. As a results, the larger the number of covered terms is, the
larger θR values our model uses and, in total, the entailment probability increases.

The 5th Recognizing Textual Entailment challenge (RTE-5) introduced a new

pilot task (Bentivogli et al., 2009) which became the main task in RTE-6
(Bentivogli et al., 2010). In this task the goal is to find all sentences that entail each
hypothesis in a given document cluster. This task’s data sets reflect a natural
distribution of entailments in a corpus and demonstrate a more realistic scenario
than the earlier RTE challenges. As reviewed in the following paragraphs there are
several characteristic in common to most entailment systems: (1) lexical resources
have a minimal impact on their performance, (2) they heuristically utilize lexical
resources, and (3) there is no principled method for making the final entailment
decision. The best performing system of RTE-5 was presented by Mirkin et. al
(2009a). It applies supervised classifiers over a parse tree representations to
identify entailment. They reported that utilizing lexical resources only slightly
improved their performance.
 MacKinlay and Baldwin (2009) presented the best lexical-level system at
RTE-5. They use a vector space method to measure the lexical overlap between the
text and the hypothesis. Since usually texts of RTE are longer than their
corresponding hypotheses, the standard cosine similarity score came lower than
expected. To overcome this problem they suggested a simple ad-hoc variant of the
cosine similarity score which removed from the text all terms which did not appear
in the corresponding hypothesis. While this heuristic improved performance
considerably, they reported a decrease in performance when utilizing synonym and
derivation relations from WordNet. On the RTE-6 data set, the syntactic-based
system of Jia et. al (2010) achieved the best results, only slightly higher than the
lexical-level system of (Majumdar and Bhattacharyya, 2010). The latter utilized
several resources for matching hypothesis terms with text terms: WordNet,
VerbOcean (Chklovski and Pantel, 2004), utilizing two of its relations, as well as
an acronym database, number matching module, co-reference resolution and
named entity recognition tools.

3. A common definition of entailment in formal semantics specifies that a text entails

another text (hypothesis, in our terminology) if h is true in every circumstance
(possible world) in which is true. In model properties, textual entailment is defined
as a relationship between texts and propositions whose representation is typically
based on text as well, unlike logical entailment which is a relationship between
propositions only. Accordingly, textual entailment confidence is conditioned on the
actual generation of a text, rather than its truth.

For illustration, we would expect that the text “His father was born in Italy”
would logically entail the hypothesis “He was born in Italy” with high probability -
since most people who’s father was born in Italy were also born there. However we
expect that the text would actually not probabilistically textually entail the
hypothesis since most people for whom it is specifically reported that their father
was born in Italy were not born in Italy.

Bar-Ilan system participating in the First Recognising Textual Entailment

Challenge. We proposed a general probabilistic setting that formalizes the notion
of textual entailment. In addition we described an alignment-based model in a bag
of words representation for lexical entailment, which was applied using web co-
occurrence statistics. Although our proposed lexical system is relatively simple, as
it does not rely on syntactic or other deeper analysis, it nevertheless achieved
competitive results. These results may suggest that the proposed probabilistic
framework is a promising basis for improved implementations that would
incorporate deeper types of information.

4. This journal tells about two ways to detect entailment using distributional
semantic representations of phrases. And this journal explain about the entailment
relation between adjective-noun constructions and their head nouns, once
represented as semantic vector pairs, generalizes to lexical entailment among
nouns dog |= animal and this journal shows that a classifier fed semantic vector
pairs can similarly generalize the entailment relation among quantifier phrases
many dogs |= some dogs to entailment involving unseen quantifiers all cats |=
several cats.
Entailment in formal semantics to characterize the conditions under which a
sentence is true, FS ( formal semantic ) begins with the lexical meanings of the
words in the sentence and builds up the meanings of larger and larger phrases until
it arrives at the meaning of the whole sentence. The meanings throughout this
compositional process inhabit a variety of semantic domains, depending on the
syntactic category of the expressions: typically, a sentence denotes a truth value (
true or false ) or truth conditions, a noun such as cat denotes a set of entities, and a
quantifier phrase ( QP ) such as all cats denotes a set of sets of entities.

The experiments show that semantic vectors contain enough information to

detect a logical relation such as entailment not only between words, but also
between phrases containing quantifiers that determine their entailment relation.
 References
Vulic. et al (2017). Hyperlex: A Large-Scale Evaluation of Graded Lexical
Entailment. The MIT Press Journals. (Accessed 15/03/2018)

Glickman. Et al. A Lexical Alignment Model For Probablistic Textual Alignment.

Shnarch. Et al. Towards a Probabilistic Model for Lexical Entailment.

Baroni. Et al. Entailment above the word level in distributional semantics