Advances in phonetics-based sub-unit modeling for transcription alignment and sign language recognition
Summary (2 min read)
1. Introduction
- The concept of an “Asian premium,” which originated in crude oil markets, has been extended to the natural gas market.
- Therefore, prices on the three markets may not be comparable.
- This is because if the market fundamentals are the determining factors of natural gas prices in East Asia, moving away from oil indexation may not be the solution to the Asian premium.
- Using historical data from three countries—Japan, the United States, and Germany—the authors hope to reveal the role that various factors—supply, demand, global economic conditions, and the oil market—play in natural gas price variation and how their roles change over time.
- The remainder of this paper is organized as follows.
2. Literature review
- The Asian premium in the gas sector has been debated in both academia and the gas industry, generated by a real policy issue over whether to retain the oil-indexed gas pricing mechanism.
- Many studies observe that natural gas prices used to link to oil prices but decoupled recently, and they were much more volatile than oil prices (Geng et al., 2016b; Serletis and Shahmoradi, 2005).
- One key question in the debate on the Asian premium is that whether it represents price discrimination or simply reflects differences in fundamentals among different markets.
- And this is driven by the weather factor, indicating the important role of market fundamentals.
- This paper tries to fill this gap by revealing the driving factors behind gas pricing in East Asia (represented by Japan) and the other two major markets (the United States and Europe).
3. Methodology
- Following the literature, this paper constructs a system based on the VAR approach introduced by Diebold and Yilmaz (2009).
- Consider a VAR model of the following form: 1 p t i t i t i yy (1) where t is a vector of independently identically distributed disturbances, and s are matrices of coefficients to be estimated.
- The net directional connectedness (NDC) is therefore H Hi iC C - .
- These two data series are collected from the International Monetary Fund Primary Commodity Prices.
- Among the early similarities in pricing patterns in Japan and Germany are that natural gas prices are largely indexed to oil prices in these markets.
5. Empirical results
- A seven-variable VAR model is estimated for each country.
- The empirical results for the full sample and rolling-windows estimation are reported in sections 5.1 and 5.2.
- First, oil price changes contribute significantly to the system for Japan and Germany, 62.77% and 50.26%, respectively, whereas it only offers 23.13% to the system for the United States.
- Oil price changes also have very low share of influence (only 5.18% and ranking number 3 in all five factors) in the US market.
- This finding is consistent with the previous literature (e.g., Ji et al., 2014).
5.2 Rolling-windows estimation
- Over the full sample period, market conditions have continually changed.
- One would expect to see that the structure/mechanism of this system may be time varying.
- Overall, oil’s contribution to gas prices in Japan and Germany shows a declining trend in recent years, which is to be expected.
- Four lags for each rolling window are used to make the results comparable.
5.3 Subsample analysis
- Rolling-windows analysis shows that the system may experience structural changes.
- This is not the focus of their paper.
- The full sample analysis shows that oil price changes are the most important contributor to the dynamics of natural gas prices in Japan and Germany.
- The impacts from market fundamentals and global economic conditions differ from country to country.
- 24 Given the recent trend toward the financialization of energy markets, natural gas prices are expected to respond less to market fundamentals and more to financial markets and trading mechanisms.
Did you find this useful? Give us your feedback
Citations
309 citations
Cites background from "Advances in phonetics-based sub-uni..."
...[55] extract sub-unit definitions from linguistic annotation in HamNoSys [31] to improve an HMM-based system recognising...
[...]
146 citations
135 citations
Cites background from "Advances in phonetics-based sub-uni..."
...Future Work The learnt sub-units show promise and, as shown by the work of Pitsikalis et al. (2011), there are several avenues which can be explored....
[...]
...Using the motion of the hands, the sign can be split into its component parts (as in Pitsikalis et al., 2011), that are then aligned with the sign annotations....
[...]
...The learnt sub-units show promise and, as shown by the work of Pitsikalis et al. (2011), there are several avenues which can be explored....
[...]
...Cooper and Bowden (2010) learnt linguistic sub-units from hand annotated data which they combined with Markov models to create sign level classifiers, while Pitsikalis et al. (2011) presented a method which incorporated phonetic transcriptions into sub-unit based statistical models....
[...]
74 citations
Cites methods from "Advances in phonetics-based sub-uni..."
...[9] proposed a method which uses linguistic labelling to split signs into sub-units....
[...]
72 citations
Cites methods from "Advances in phonetics-based sub-uni..."
...…recognition from these sequential observations, hidden state–based methods like the hidden Markov model (HMM) [Liang and Ouhyoung 1996; Starner et al. 1998; Gao et al. 2004; Pitsikalis et al. 2011] and conditional random fields (CRF) [Yang et al. 2009; Kong and Ranganath 2014] were frequently used....
[...]
References
703 citations
"Advances in phonetics-based sub-uni..." refers background or methods or result in this paper
...In addition, we consider epenthesis movements (E) [7] to be distinct from T; the former are transitions between two locations without an explicit path, and primarily occur when the hands move into position between signs, and during repeated movements....
[...]
...This also is consistent with the concepts in the old Movement-Hold model [7]....
[...]
...It supersedes the older Movement-Hold model [7] used in earlier work, and fixes many of its shortcomings(1)....
[...]
321 citations
"Advances in phonetics-based sub-uni..." refers background in this paper
...There has been little progress in the area of phonetic modeling for the purposes of SL recognition since the work of Vogler and Metaxas [11]....
[...]
...Like in the work by Vogler and Metaxas, the basic phonetic structure of a sign is a sequence of segments, which we model according to Johnson’s and Liddell’s recent work on the Posture-Detention-Transition-Steady Shift (PDTS) system [6]....
[...]
178 citations
"Advances in phonetics-based sub-uni..." refers background in this paper
...Recent successful data-driven methods include [1, 4, 2, 5, 3, 12, 8]....
[...]
...One employs a linguistic feature vector based on measured visual features, such as relative hand movements [2]....
[...]
101 citations
81 citations
"Advances in phonetics-based sub-uni..." refers background in this paper
...Instead of single frames, [4, 5, 12] cluster sequences of frames on the feature level, such that they exploit the dynamics inherent to sign language....
[...]
...Recent successful data-driven methods include [1, 4, 2, 5, 3, 12, 8]....
[...]
...Other previous approaches include [1, 4, 5]....
[...]
Related Papers (5)
Frequently Asked Questions (8)
Q2. What are the causes of outliers and high variances?
Outliers and high variances seem to be caused by visual processing inaccuracies (we perform 2D, rather than 3D, processing), tracking or parameter estimation errors, or human annotator errors, or actual data exhibiting such properties.
Q3. What are the steps involved in the conversion of the phonetic labels?
The procedures involved in this process involve: (1) phonetic sub-unit construction and training, (2) phonetic label alignment and segmentation, (3) lexicon construction, and (4) recognition.
Q4. how can other disciplines benefit from their results for the analysis of sign languages?
the authors expect that other disciplines, such as linguistics, can greatly benefit from their results for the analysis of sign languages.
Q5. What is the morphological processing used for the detection of the signer’s hands and?
For the segmentation and detection of the signer’s hands and head in the Greek Sign Language (GSL) Lemmas Corpus, the authors employed a skin color model utilizing a Gaussian Markov Model (GMM), accompanied by morphological processing to enhance skin detection.
Q6. What is the meaning of the annotations?
The annotations of the signs are coded in HamNoSys [9], a symbolic annotation system that can describe a sign in sufficient detail to display it in an animated avatar.
Q7. What is the expected effect of increasing the number of signs?
By increasing the number of signs, the recognition performance for both approaches decreases; this is expected as the recognition task becomes harder.
Q8. What is the conversion method for PILE?
Their conversion method from HamNoSys to the PDTS structure resolves the implied parts, and splits the signs into its constituent segments.