Title: The Role of CNL and AMR in Scalable Abstractive Summarization for Multilingual Media Monitoring
ArXiv ID: 1606.05994
Date: 2016-07-19
Authors: Normunds Gruzitis and Guntis Barzdins
📝 Abstract
In the era of Big Data and Deep Learning, there is a common view that machine learning approaches are the only way to cope with the robust and scalable information extraction and summarization. It has been recently proposed that the CNL approach could be scaled up, building on the concept of embedded CNL and, thus, allowing for CNL-based information extraction from e.g. normative or medical texts that are rather controlled by nature but still infringe the boundaries of CNL. Although it is arguable if CNL can be exploited to approach the robust wide-coverage semantic parsing for use cases like media monitoring, its potential becomes much more obvious in the opposite direction: generation of story highlights from the summarized AMR graphs, which is in the focus of this position paper.
💡 Deep Analysis
📄 Full Content
The Role of CNL and AMR in Scalable Abstractive
Summarization for Multilingual Media Monitoring
Normunds Gruzitis and Guntis Barzdins
University of Latvia, IMCS and LETA
normunds.gruzitis@lumii.lv, guntis.barzdins@leta.lv
In the era of Big Data and Deep Learning, a common view is that statistical and ma-
chine learning approaches are the only way to cope with the robust and scalable in-
formation extraction and summarization. Manning [1] compares Deep Learning with
a tsunami at the shores of Computational Linguistics, raising a question if this is the
end for the linguistically oriented approaches. Consequently, this question is relevant
also to the special interest group on Controlled Natural Language (CNL).
It has been recently proposed that the CNL approach could be scaled up, building
on the concept of embedded CNL [2] and, thus, allowing for CNL-based information
extraction from e.g. normative or medical texts that are rather controlled by nature but
still infringe the boundaries of CNL or the target formalism [3]. It has also been
demonstrated that CNL can serve as an efficient and user-friendly interface for Big
Data end-point querying [4; 5], or for bootstrapping robust NL interfaces [6]. as well
as for tailored multilingual natural language generation from the retrieved data [4].
In this position paper, we focus on the issue of multi-document storyline summari-
zation, and generation of story highlights – a task in the Horizon 2020 Big Data pro-
ject SUMMA1 (Scalable Understanding of Multilingual MediA). For this use case, the
information extraction process, i.e., the semantic parsing of input texts cannot be ap-
proached by CNL: large-scale media monitoring is not limited to a particular domain,
and the input sources vary from newswire texts to radio and TV transcripts to user-
generated content in social networks. Robust machine learning techniques are neces-
sary instead to map the arbitrary input sentences to their meaning representation in
terms of PropBank and FrameNet [7], or the emerging Abstract Meaning Representa-
tion, AMR [8], which is based on PropBank with named entity recognition and link-
ing via DBpedia [9]. AMR parsing has reached 67% accuracy (the F1 score) on open-
domain texts, which is a level acceptable for automatic summarization [10].
Although it is arguable if CNL can be exploited to approach the robust wide-
coverage semantic parsing for use cases like media monitoring, its potential becomes
much more obvious in the opposite direction: generation of story highlights from the
summarized (pruned) AMR graphs. An example of possible input and expected out-
put is given in Figure 1.
While novel methods for AMR-based abstractive 2 summarization begin to ap-
pear [11], full text generation from AMR is still recognized as a future task [11],
1 http://summa-project.eu
2 Abstractive summarization contrasts extractive summarization which selects representative
sentences from the input documents, optionally compressing several sentences into one.
which is an unexplored niche for the CNL and grammar-based approaches.3 Here we
see, for instance, Grammatical Framework, GF [12], as an excellent opportunity for
implementing an AMR to text generator.
Article1
[..] An ongoing battle in Aleppo eventually terminated when the rebels took over the
city. [..] President Assad gave a speech, denouncing the death of soldiers. [..]
Article2
[..] Syrian rebels took control of Aleppo. [..]
Article3
[..] The Syrian opposition forces won the battle over Aleppo city. [..] Syrian president
announced that such insurgence will not be tolerated. [..]
Output Summary:
Syrian rebels took over Aleppo
Article1 Article2 Article3
Assad gave a speech about the battle
Article1 Article3
Fig. 1. Abstractive summarization. An example from the SUMMA proposal
The summarized AMR graphs would have to be mapped to the abstract syntax trees
(AST) in GF (see an example in Figure 2). As GF abstract syntax can be equipped
with multiple concrete syntaxes, reusing the readily available GF resource grammar
library, this would allow for multilingual summary generation, even extending the
SUMMA proposal.
Abstract Meaning Representation (AMR)
GF Abstract Syntax Tree (AST)
(x3 / take-01
:ARG0 (x2 / organization
:wiki “Syrian_opposition”
:name (n2 / name
:op1 “Syrian” :op2 “rebels”))
:ARG1 (x5 / city :wiki “Aleppo”
:name (n / name :op1 “Aleppo”)))
PredVP
organization_Syrian_opposition_NP
(ComplV2 take_over_V2
city_Aleppo_NP)
Fig. 2. AMR and AST representations of “Syrian rebels took over Aleppo”
We assume that the generation of story highlights in the open newswire domain is
based on a relatively limited set of possible situations (semantic frames) and a rela-
tively limited set of syntactic constructions (a restricted style of writing) similar to the
Multilingual Headlines Generat
